From 82c0a8e61fa5498c7fd67d54772ed7a94f7673ea Mon Sep 17 00:00:00 2001 From: Claude Date: Sun, 30 Aug 2026 18:53:47 +0000 Subject: [PATCH 1/9] Session arrival: Paper A's revision, carrying W1 (2026-08-30) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The session's first act, before any repository is read for the work: prompt.md and bootstrap.md land under meta/sessions/2026-08-30-paper-a-revision/, with the package README under inputs/ and the sessions index extended. Five of the seven package files are already committed byte-identically — the review and the triage at 2026-08-23-phase1a/, and Q44, Q45 and the ontology explainer at 2026-08-27-ground-migration/inputs/. Each was hashed against its committed copy before the decision not to re-file, and those paths are the ones this session cites. Filing Q44 and Q45 nowhere new is deliberate: they are unfiled downstream demand, the prompt keeps them unfiled, and the paper may not cite them. Base commits recorded in bootstrap.md: actor-indexed-determination at 81f6929, which annotated tag v5.12.0 resolves to exactly; this repository at 54f00eb. Basis: DDD-dec-20; DDD-dec-17; DDD-delivery-02; term:delivery --- .../2026-08-30-paper-a-revision/bootstrap.md | 101 ++++++++++++++ .../inputs/README.md | 45 ++++++ .../2026-08-30-paper-a-revision/prompt.md | 130 ++++++++++++++++++ meta/sessions/README.md | 1 + 4 files changed, 277 insertions(+) create mode 100644 meta/sessions/2026-08-30-paper-a-revision/bootstrap.md create mode 100644 meta/sessions/2026-08-30-paper-a-revision/inputs/README.md create mode 100644 meta/sessions/2026-08-30-paper-a-revision/prompt.md diff --git a/meta/sessions/2026-08-30-paper-a-revision/bootstrap.md b/meta/sessions/2026-08-30-paper-a-revision/bootstrap.md new file mode 100644 index 0000000..7cc0061 --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/bootstrap.md @@ -0,0 +1,101 @@ +# Bootstrap — Paper A's revision, carrying W1 (2026-08-30) + +The invocation message that accompanied `prompt.md`, recorded verbatim as the session's arrival +record. Committed **before** any repository is read for the work, per `DDD-dec-20`. + +## Parameters + +- **Branch:** `claude/paper-a-w1-revision-b8sx45` (both repositories; the branch name was fixed by + the invocation harness and is mirrored across the two so the set is one name) +- **Base commits:** + - `actor-indexed-determination` — head `81f6929d7525bcb1f2d07b5ce5bf3c6ed6d4275d`. Annotated tag + `v5.12.0` resolves to **the same commit** (`git rev-parse 'v5.12.0^{}'` → + `81f6929d7525bcb1f2d07b5ce5bf3c6ed6d4275d`); head and tag coincide, and canon at head is canon + at `v5.12.0` exactly as the prompt asserts. Tag message: *"v5.12.0 — Definitions that were + never in the registry, and a collision the store won"*. This is the tag the pin advances to; + the projection's pin at arrival is `v5.9.0`. + - `decision-driven-design` — head `54f00eb775edf75fe37649b9a99e8443a014ee8f` (the item-5 merge, + PR #31) +- **Gates:** 5 (G1 survey plan, retitle, supplement boundary, W1's counts · G2 the related-work + section · G3 the narrowed claim, the rhetoric pass, the small items · G4 pin advance, + quotations, Appendix A, supplement split, W1 · G5 close) +- **Principal:** Emil +- **Session type:** interactive paper revision — projection work; hold at every gate, merge nothing +- **Input identity:** + + | File | Lines | sha256 | + |---|---|---| + | `prompt.md` | 130 | `6a0ca35825259ac09c4357d2561be069f3b5dcc89fbd71399c0c9d24239e975c` | + | package `README.md` | 45 | `8808bebae00354acfa37adca92b0860a804b9fe16244af467c4137052e1c2588` | + | package `paper-a-objective-review.md` | 349 | `1d291a5fcc605ae6ade4c82922802b1be2d132d33b470110946d95b272205707` | + | package `paper-a-review-triage.md` | 179 | `e9650358e5f6120905d4d7cfc4f4769c407d6f10d4804be9fe09bada8e3f741a` | + | package `q44-act-and-verdict-ontology.md` | 212 | `4d49ded8c34ac0a48b096040932ca478eae5140ad5aa26f16302f2171a740af8` | + | package `q45-routes-and-compositional-coverage.md` | 114 | `c5925b25656a53279f858b742f3cecd84acd2052ea7d2ead9ccfd5f0a52c051b` | + | package `act-and-verdict-ontology-explainer.md` | 222 | `644eabd6a70c74894ae195c3ddda72728932562786e974479041b329b5fe59b5` | + +## Arrival — clean + +The prompt arrived with the invocation, inside `papera-package.zip`, and is filed here with its +identity in this session's first commit. No supplementary input was named and none is missing. + +**Five of the seven package files are already committed, byte-identically, and are not re-filed.** +Each was hashed against its committed copy before the decision: + +| Package file | Committed at | Identical | +|---|---|---| +| `paper-a-objective-review.md` | `meta/sessions/2026-08-23-phase1a/paper-a-objective-review.md` | yes | +| `paper-a-review-triage.md` | `meta/sessions/2026-08-23-phase1a/paper-a-review-triage.md` | yes | +| `q44-act-and-verdict-ontology.md` | `meta/sessions/2026-08-27-ground-migration/inputs/q44-act-and-verdict-ontology.md` | yes | +| `q45-routes-and-compositional-coverage.md` | `meta/sessions/2026-08-27-ground-migration/inputs/q45-routes-and-compositional-coverage.md` | yes | +| `act-and-verdict-ontology-explainer.md` | `meta/sessions/2026-08-27-ground-migration/inputs/act-and-verdict-ontology-explainer.md` | yes | + +Those five paths are the ones cited throughout this session. The identity check is the point, not +the courtesy: the ground-migration session's arrival caught a wrong artefact offered at a gate by +hashing it, and an unhashed "already committed" would be the same presumed discharge in a quieter +form. + +The two files with no committed copy are `prompt.md` (this session's charter) and the package +`README.md`, filed under `inputs/` so the charter is reconstructible. + +**Filing Q44 and Q45 anywhere is not filing them as canon.** They are unfiled downstream demand and +the prompt keeps them unfiled. This session rules *with their existence known* — which is only +possible if they are quotable at the gates — and the paper may not cite them. Where the paper wants +a distinction they carry, it is register-native prose or it waits. + +## The invocation message, verbatim + +Unzip papera-package.zip and read papera-package/README.md, then read prompt-paper-a-revision.md in +its entirety — this session follows that prompt exactly, including every gate. + +This is Paper A's revision: it answers its external review, advances its pin from v5.9.0 to v5.12.0, +and carries W1 — the ground migration's cheapest wave, 88% of which lives inside this paper and the +measure note. + +First act, before anything else: commit this prompt and bootstrap to +meta/sessions/2026-08-30-paper-a-revision/ in decision-driven-design, per DDD-dec-20. + +Fetch both repos at head — actor-indexed-determination at v5.12.0 (verify and report the tag found) +and decision-driven-design at head. The manuscript at papers/paper-a/paper-a.md is the working text. + +Rules that override anything you might infer: +- Interactive revision. Stop at every gate for Emil's ruling. Merge nothing. +- Projection work: the paper may not introduce claims. Q44, Q45 and the ontology explainer are + supplied as context and are UNFILED — never cite them as canon. Where the paper wants a + distinction they carry, it is register-native prose or it waits. +- The related-work survey is reading before it is drafting. Verify every locator or flag it + unverified in its own entry; the Hayek facsimile check and the Tesler no-primary-publication entry + are the two patterns. +- W1 renames the population sense ONLY. Any occurrence ambiguous between it and another sense defers + to the migration rather than being ruled here. Report moved and deferred counts. +- State predicted W6/W7 results before the pin advance, and verify after. +- Appendix A is regenerated wholesale from the graph, never hand-edited, with the independent + re-read. Hand-editing it would forge agreement the pin does not have. +- Commit drafts before reporting at each gate, bodies marked draft-pending-ruling. + +The failure mode to avoid: the revision makes the paper shorter, better sourced and narrower, and a +shorter paper can end up sounding more certain because it has less to qualify. Where a claim is +projected, the sentence around it says so. + +Begin with step 1 and end your first report at GATE 1: citations re-verified at v5.12.0, the four +failing quotations confirmed, W1's occurrences enumerated with ambiguous cases separated, the survey +plan, retitle candidates, and the supplement split's proposed boundary. diff --git a/meta/sessions/2026-08-30-paper-a-revision/inputs/README.md b/meta/sessions/2026-08-30-paper-a-revision/inputs/README.md new file mode 100644 index 0000000..3c35d0e --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/inputs/README.md @@ -0,0 +1,45 @@ +# Paper A's revision — package + +Upload to a Claude Code session with both canon repositories available, and paste `bootstrap.md` as +the opening message. + +## Contents + +| File | Role | +|---|---| +| `prompt-paper-a-revision.md` | The session prompt. Five gates, six work items | +| `bootstrap.md` | Paste as the opening message | +| `paper-a-objective-review.md` | The external review — the session's brief | +| `paper-a-review-triage.md` | Emil's ruled triage: concede, defend, repair | +| `q44-act-and-verdict-ontology.md` | **Unfiled.** Context for differentiating in the survey | +| `q45-routes-and-compositional-coverage.md` | **Unfiled.** Same | +| `act-and-verdict-ontology-explainer.md` | **Unfiled.** Standalone explanation of both | + +**Not in the package, because the repositories are authoritative:** the manuscript +(`papers/paper-a/paper-a.md`), Appendix A's generator and checkers, `measure-paper-context.md`, and +the migration seed (`meta/migration-plan-ground.md`). + +## Why this runs before the migration + +W1 is not a dependency of the revision — it is nearly the same work. 88 of its 88 mutable +occurrences sit in the two papers; 11 more ride other waves trivially. And W1 is separable by +construction: it renames a sense that touches no other sense and fires no pins, so carrying it here +is not a partial migration. + +The write-twice cost of doing the papers first is 13 occurrences of the migration's 187 — and zero +if W1 rides, which it does. + +## What Emil rules + +| Gate | Ruling | +|---|---| +| 1 | Survey plan, retitle candidates, supplement boundary, W1's counts | +| 2 | The related-work section — the session's largest ruling | +| 3 | The narrowed novelty claim, the rhetoric pass, the review's small items | +| 4 | Pin advance, the four quotations, Appendix A, the supplement split, W1 | +| 5 | Close, PR, the response-to-review document | + +## After this + +The ground migration executes from its seed — now 88 occurrences smaller. Then the primer, then the +transfer. diff --git a/meta/sessions/2026-08-30-paper-a-revision/prompt.md b/meta/sessions/2026-08-30-paper-a-revision/prompt.md new file mode 100644 index 0000000..85bf4a4 --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/prompt.md @@ -0,0 +1,130 @@ +# Session prompt — Paper A's revision, carrying W1 + +Repositories: `actor-indexed-determination` (upstream, head — canon at **v5.12.0**; verify) and +`decision-driven-design` (downstream, head — the manuscript's home). Fetch both. +Session type: **interactive paper revision.** Projection work: the paper may not introduce claims; +every load-bearing statement cites a graph node or is flagged register-native. Hold at every gate. +British spelling; one idea per sentence; principle, never law. +**First act, before Gate 1:** commit this prompt and bootstrap to +`meta/sessions/2026-08-30-paper-a-revision/` in `decision-driven-design`, per DDD-dec-20. + +## What this session is + +Paper A answers its external review, advances its pin from v5.9.0 to **v5.12.0**, and **carries W1** +— the migration's cheapest wave, 88% of which lives inside this paper and the measure note. + +**The review is a major-revision review and it is correct on most counts.** Where the triage says +concede, concede cleanly; where it says defend, defend with the framework's own material rather than +with rhetoric. The paper's credibility has come entirely from conceding before it was caught. + +**Scope note on W1:** it is separable by construction — S5 leaves the word entirely, its replacement +name is settled, it fires no pins, and it touches no S1/S2/S3 occurrence. Carrying it here is **not** +a partial migration, which is the only thing the migration's defer-whole ruling protects. + +## The work, itemised + +### R-1 — the related-work survey *(the largest item, and it is reading)* + +The review names seven works the paper does not engage. Each needs enough engagement to say **what +the framework takes and where it differs** — the standard the measure note's §8 set, not a citation +list: + +- **Hutchins**, distributed cognition +- **Hollnagel & Woods**, joint cognitive systems — **closest to the arrangement**, and the one that + most narrows the novelty claim +- **Leveson**, STAMP — control structure, but not the supply partition +- **Horvitz**, mixed-initiative +- **Bovens**, accountability +- **Matthias**, the responsibility gap +- **Meaningful human control**, as a literature + +**Verify every locator or flag it unverified in its own entry** — the Hayek precedent stands: the +facsimile check caught two errors a database would have propagated, and the Tesler entry is the +pattern for a source with no primary publication. + +**Sharper ground to differentiate on than the review had.** Q44 and Q45 (supplied) give the paper +distinctions the reviewer did not have: Hollnagel & Woods have joint cognitive systems but not the +supply partition; Leveson has control structure but not filing-versus-delivery. **The framework's +genuine difference is more statable now than when the review was written** — but the survey must be +read before the difference is claimed, not after. + +**Do not cite Q44 or Q45 as canon. They are unfiled.** Where the paper wants a distinction they +carry, it is register-native prose or it waits. + +### R-2 — the narrowed novelty claim, the retitle, and the honest scope + +- Adopt the reviewer's narrowed claim in substance: **a specific, auditable synthesis of resolution, + assurance, delivery and accountability** — not the discovery of the arrangement. +- **Retitle.** *The Missing Parameter* claims absence from prior work the paper has not surveyed. + Propose alternatives at Gate 1 with the survey's shape in view. +- §11's rhetoric pass: the review's finding was that the surrounding sentences overrun the status + labels. Do not let a shorter paper become a more confident one. + +### R-3 — the four failing quotations and the pin advance + +The pin was held at v5.9.0 by ruling so canon could move without dragging the projection. It advances +here, to **v5.12.0**, and four quotations fail against it: + +- `DDD-frame-15`'s statement (retired; superseded by `DDD-frame-17`) +- `DDD-frame-02`'s statement (amended) +- `term:residual-discretion`'s canonical text (amended) +- the compact form quoted inline at ~L1208, disclosed-partial — **the cheapest of the four: the + compact form survives verbatim in `DDD-frame-17`, so the citation moves and no prose does** + +Two of these are the very claims the review asks the paper to stop overstating, so the repair pays a +debt it already owed. Appendix A regenerates wholesale against the new pin, with the independent +re-read. + +### R-4 — the supplement split + +Move Appendix A and the projection mechanics out of the argumentative path. The review's finding: an +archival paper should not carry filing history, pending-node discussion, or graph machinery in its +main body. **The apparatus stays — it moves.** + +### R-5 — W1, riding + +`deployment distribution` replaces the population sense of the word throughout the paper and, where +the same session can reach it, the measure note. 88 occurrences of W1 sit in these two artefacts. + +**Constraints:** this renames **only** the population sense. It touches no S1, S2 or S3 occurrence, +and if a candidate occurrence is ambiguous between S5 and any other sense, it **defers to the +migration** rather than being ruled here. Report the count moved and the count deferred. + +### R-6 — the review's remaining items + +From the triage, each small: §9's hypotheses as separate preregistered studies with **coding +reliability first**; §10.2's accountability element counts reconciled (five versus three); §10.3's +provenance classifications restated as hypotheses rather than facts supplied by the scenario; and +§5's residual-discretion separation reflected where the paper uses the term. + +## Walk + +1. **Fetch, inaugurate, survey.** Both repos at head; session record committed; every citation + re-verified at v5.12.0; the four failing quotations confirmed; W1's candidate occurrences + enumerated with ambiguous cases separated. Report: the survey plan for R-1, retitle candidates, + the supplement split's proposed boundary, and W1's counts. **GATE 1 — hold.** +2. **R-1 drafted** — the related-work section, with every locator verified or flagged. + **GATE 2 — hold; this is the session's largest ruling.** +3. **R-2 and R-6.** The narrowed claim, the retitle, the rhetoric pass, and the review's small items. + **GATE 3 — hold.** +4. **R-3, R-4, R-5.** Pin advance with predicted W6/W7 stated before the operation; the four + quotations; Appendix A regenerated and independently re-read; the supplement split; W1. + **GATE 4 — hold.** +5. **Close.** Full read; validators; reference closure; quotation and appendix checkers; word count + with method stated; branch, PR, manifest; a response-to-review document mapping every objection to + conceded, repaired, or defended-and-why. **GATE 5 — hold.** + +## Out of scope + +Filing anything in canon — including Q44, Q45 and Q46, which are unfiled and stay unfiled. The +migration proper (its seed is at `meta/migration-plan-ground.md`). The primer. The transfer. W4. The +decoder repair. The `verdict` and `projection` collisions, now freight-listed with their contract +findings. Do not bundle. + +## Standing note + +Commit drafts before reporting at each gate, bodies marked draft-pending-ruling. The review's +sharpest structural criticism was that the paper communicates more warrant than the graph provides. +The revision makes the paper shorter, better sourced, and narrower — and the failure mode is a +shorter paper that sounds more certain because it has less to qualify. Where a claim is `projected`, +the sentence around it says so. diff --git a/meta/sessions/README.md b/meta/sessions/README.md index 6bfc704..54f289b 100644 --- a/meta/sessions/README.md +++ b/meta/sessions/README.md @@ -34,3 +34,4 @@ session's later work exceeds the committed prompt, the excess is visible as exce | `2026-08-24-ground-audit/` | Interactive audit — Phase 1b: the ground audit (D-1…D-4); changed nothing. Output: `meta/ground-audit-2026-08-24.md` | | `2026-08-25-item4/` | Interactive canon curation — item 4: status, kind, the validator and the appendix (I-1…I-4) | | `2026-08-27-ground-migration/` | Interactive canon curation with an execution phase — item 5: the ground migration, on the ground audit's rulings (G1…G6) | +| `2026-08-30-paper-a-revision/` | Interactive paper revision — Paper A answers its external review, advances its pin to `v5.12.0`, and carries W1 (R-1…R-6) | From 553abbd66a99c1bea523d53c0470bbe7c73a1e6e Mon Sep 17 00:00:00 2001 From: Claude Date: Sun, 30 Aug 2026 19:06:31 +0000 Subject: [PATCH 2/9] GATE 1 (draft-pending-ruling): citations at v5.12.0, W1's occurrences, three plans MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Nothing in papers/ is touched. No claim, term or decision is filed, amended or retired in either repository, and the pin stays at v5.9.0. The whole of this diff is the gate report and its two instruments. CITATIONS. All 72 nodes Paper A cites still resolve at v5.12.0 — no E12-class loss. check-quotations.py reports exactly the four failing quotations the prompt predicted (DDD-frame-15's statement, DDD-frame-02's, term:residual- discretion's, and the compact form at L1208, whose citation moves to DDD-frame-17 while its prose does not). check-appendix.py reports ten discrepancies across eight nodes, all content movement. A FIFTH REPAIR NO CHECKER SEES. DDD-measure-06 is retired at v5.12.0, superseded by DDD-measure-16 and DDD-measure-17 — the review's F-A, already repaired in canon. The paper asserts it inline as **established** three times and neither existing checker reads an inline status assertion. status-sweep.py closes that gap: 68 assertions checked, 0 wrong at v5.9.0, 4 wrong at v5.12.0. W1. 18 proposed moved, 13 deferred, of the 31 S5 occurrences in the two manuscripts. The deferrals are enumerated line by line with a reason each: generated appendix rows, passages quoting a live claim verbatim, and one filename. The ambiguous "ground the task faces" pair is separated and flagged rather than ruled, per the migration's own constraint. A COUNT IN THE CHARTER THAT DOES NOT RECONCILE. 88 is the corpus-wide mutable S5 total, not the count in the two papers. They hold 31 of it — 35%, not 88%. Reported rather than worked around; the close report will say 31 of 88 and the migration's remaining W1 surface is 57. Basis: DDD-dec-20; DDD-agent-01; DDD-delivery-02; term:escape; DDD-frame-15; DDD-frame-17; DDD-measure-06; DDD-measure-16; DDD-measure-17; DDD-frame-02; term:residual-discretion; DDD-measure-12; term:verdict --- .../gate1-survey.md | 391 ++++++++++++++++++ .../status-sweep.py | 91 ++++ .../w1-enumerate.py | 103 +++++ 3 files changed, 585 insertions(+) create mode 100644 meta/sessions/2026-08-30-paper-a-revision/gate1-survey.md create mode 100755 meta/sessions/2026-08-30-paper-a-revision/status-sweep.py create mode 100644 meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py diff --git a/meta/sessions/2026-08-30-paper-a-revision/gate1-survey.md b/meta/sessions/2026-08-30-paper-a-revision/gate1-survey.md new file mode 100644 index 0000000..a068328 --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/gate1-survey.md @@ -0,0 +1,391 @@ +# GATE 1 — citations at `v5.12.0`, W1's occurrences, and three plans + +**Status: draft-pending-ruling.** Nothing in the manuscript is touched. This gate changed no claim, +no term and no prose; the whole of its diff is this document and the session's arrival record. + +**Read at:** `actor-indexed-determination` `81f6929` · `decision-driven-design` `54f00eb` +(both repositories fetched at head on the session's branch, `claude/paper-a-w1-revision-b8sx45`). + +**The tag, verified rather than assumed.** `git rev-parse 'v5.12.0^{}'` → +`81f6929d7525bcb1f2d07b5ce5bf3c6ed6d4275d`, which is `origin/main`'s head. Head and tag are the +**same commit**, so canon at head is canon at `v5.12.0` exactly as the prompt asserts. Tag message: +*"v5.12.0 — Definitions that were never in the registry, and a collision the store won"*. The +projection's pin at arrival is `v5.9.0` (`graph/upstream.yaml`), held there by ruling. + +--- + +## 1. Every citation re-verified at `v5.12.0` + +Run against the manuscript unchanged, at both refs, with the instruments that live beside it. + +| Instrument | at `v5.9.0` (the pin) | at `v5.12.0` (the target) | +|---|---|---| +| `check-quotations.py` | 29 verbatim, 0 disclosed-partial, **0 failing** | 25 verbatim, **4 failing** | +| `check-appendix.py` | 72 nodes, **0 discrepancies** | 72 nodes, **10 discrepancies** across 8 nodes | +| id resolution | 72/72 resolve | **72/72 resolve** — no E12-class loss | + +**Nothing the paper cites has disappeared.** Every one of the 72 nodes still exists at `v5.12.0`; +what has moved is content and status, which is the governed case rather than the broken one. + +### 1.1 The four failing quotations — confirmed, and each named + +The prompt predicted four. The checker reports exactly four, and they are the four predicted. + +| # | Site | Node | What happened upstream | Cost | +|---|---|---|---|---| +| 1 | §4.1, L366–368 | `DDD-frame-15` statement | **Retired** at `v5.10`, superseded by `DDD-frame-17` | the block is rewritten and the surrounding four modes go | +| 2 | §3, L274–276 | `DDD-frame-02` statement | **Amended** at `v5.10` — the residual-discretion clause only; first clause untouched | one clause of the block | +| 3 | §3.2, L310–314 | `term:residual-discretion` | **Amended** at `v5.10` — the separation the review asked for | the block is replaced by canon's longer text | +| 4 | §12, L1208 | `DDD-frame-15` compact form | the compact form **survives verbatim** in `DDD-frame-17`'s closing clause | **the citation moves; no prose does** | + +Item 4 is confirmed as the prompt describes it. `DDD-frame-15`'s own retirement notes say so in +canon's words: *"The compact form — demand is never unmet, only ungoverned — is unchanged and is +carried by DDD-frame-17's closing clause."* The disclosure marker `— closing clause` stays valid; +only the id changes. + +### 1.2 A fifth repair the checkers cannot see + +**`DDD-measure-06` is retired at `v5.12.0`** (`retired_from: established`), superseded by +`DDD-measure-16` (availability to an arrangement) and `DDD-measure-17` (the coincidence, at +`projected`). That is the review's §2 — finding F-A — **already repaired in canon**. + +The quotation checker does not catch it, because the paper never blockquotes `DDD-measure-06`. It +asserts the node inline, with a status label, in prose. A sweep written for this gate finds four +inline status assertions that are wrong at `v5.12.0`: + +| Site | As the paper has it | at `v5.9.0` | at `v5.12.0` | +|---|---|---|---| +| §4.1 L368 | `[DDD-frame-15, projected]` | projected | **retired** | +| §4.4 L532 | `[DDD-measure-06, **established**]` | established | **retired** | +| §5.2 L605 | `[DDD-measure-06, **established**]` | established | **retired** | +| §6.1 L756 | `[DDD-measure-06, **established**]` | established | **retired** | + +Two further sites cite `DDD-measure-06` without a status label (§1.2 L111, §8.4 L911), and the +conclusion's *"What is established"* list at L1220–1223 names *"that the measure exists exactly +where the acceptance predicate closes"* among the established four. **That sentence is the single +most exposed line in the paper**, and at `v5.12.0` it is false in the graph's own terms. + +*The instruments are sound and this is not a defect in them.* `check-quotations.py` closes the gap +its docstring claims — block quotations — and says so. What has no instrument is the **inline +status assertion**, which is the more common form in the manuscript and the one that carries the +warrant a reader takes away. Proposed at Gate 4: the sweep above becomes a third checker beside +the other two, `check-status.py`, so the next pin advance cannot leave a retired node labelled +`established` in prose. **Flagged for ruling, not built yet.** + +### 1.3 Appendix A's ten discrepancies + +All ten are content movement on nodes that still resolve; none is a missing id. They are listed so +the regeneration at Gate 4 can be checked against a prediction rather than accepted on trust. + +| Node | What moved | +|---|---| +| `DDD-frame-15` | statement **and** status (retired) | +| `DDD-measure-06` | statement **and** status (retired) | +| `DDD-frame-02` | statement — the residual-discretion clause | +| `DDD-delivery-01` | statement — `mechanical delivery` → `act-triggered delivery` | +| `term:residual-discretion` | canonical wording — the four-way separation | +| `term:verdict` | canonical wording — the `specification demand` parenthetical moved to a new `denominations:` field | +| `term:composite-actor` | canonical wording — a definition promoted above the existing line | +| `term:seam` | canonical wording — same | + +Appendix A is regenerated wholesale at Gate 4 and never hand-edited. Two rows will also **appear** +that are not in the table above, because the body will newly cite `DDD-frame-17` and +`DDD-measure-16`/`DDD-measure-17`. + +### 1.4 The validators, at head, before anything moves + +| Gate | Result | +|---|---| +| `python3 validate-core-order.py core/` | **OK** — 5 documents, 5 terms, 4 embedded, 0 errors, 0 warnings; 67 pins resolved, 0 basis-loss, 0 content-drift | +| `python3 scripts/validate-claims.py core/claims/` | **valid** — 26 claims, **6 warnings**, exactly the six `CLAUDE.md` records for `v5.11.0` | +| `python3 scripts/validate-claims.py core/decisions/ --decisions` | **valid** — 21 decisions | + +One note, reported because it is a surprise rather than a failure: running the ordering validator +over the **repository root** rather than `core/` raises four `E13` errors on `core/14-maturation.md` +and `core/16-calibration-ledger.md` (embeds of four unpinned upstream ids). CI runs it on `core/`, +which is the documented gate and is green, so this is **not a regression and not this session's**. +It is recorded here because it is the shape of an escaped decision about the instrument — a check +that passes because of where it is pointed — and it belongs on a successor list, not in this +session's scope. + +--- + +## 2. W1 — the occurrences, enumerated, with the ambiguous cases separated + +### 2.1 What W1 is, and the one rule that governs it + +W1 renames the **population sense** of *ground* — sense **S5** in the ground migration's +classification, *the population the task faces* — to **deployment distribution**. It touches no S1 +(conditions in the case), no S2 (representations the arrangement holds) and no S3 (representations +delivered at the act). **Any occurrence ambiguous between S5 and another sense defers to the +migration rather than being ruled here.** + +The classification is not re-derived. It is the ground-migration session's, execution-grade and +committed at `meta/sessions/2026-08-27-ground-migration/`, read here through +`w0-full-v2.json` against the audit's own extract. + +### 2.2 A count in the charter that does not reconcile — reported, not worked around + +The package `README.md` says *"88 of its 88 mutable occurrences sit in the two papers"*; the +invocation says *"88% of which lives inside this paper and the measure note"*; the prompt's R-5 +says *"88 occurrences of W1 sit in these two artefacts."* Against the committed classification: + +| Quantity | Count | +|---|---| +| S5 occurrences, whole corpus | 97 | +| of which immutable (session records, a cut release descriptor) | 9 | +| **S5 mutable, whole corpus** | **88** | +| S5 mutable excluding `product-cli` (W4, ruled separately) | 87 | +| **S5 mutable in the two manuscripts** | **31** (measure note 26, Paper A 5) | +| S5 mutable anywhere under `papers/` | 39 | + +**88 is the corpus-wide mutable total, not the count in the two papers.** The two manuscripts hold +**31 of 88 — 35%, not 88%.** The migration plan's own W1 row is closer and still not exact: it +records *"87 mutable, 29 in merged papers"* where the extract gives 31. + +This does not change what the session should do; W1 is still separable and still worth carrying +here. It changes what the session may **claim** to have done. The close report will say +*31 of 88*, and the migration's remaining W1 surface is **57 occurrences, not zero.** + +### 2.3 The enumeration, with dispositions + +Thirty-one occurrences, each read in its own file at head. **Not every S5 occurrence is W1's to +move**, and the exclusions are the substance of this section. + +**Paper A — 5 occurrences** + +| Line | Text | Disposition | +|---|---|---| +| 502 | §4.4 *"over the distribution of ground the task faces"* | **move** — flagged, §2.4 | +| 672 | §5.3 *"the verdict function and the ground distribution"* | **move** | +| 1376 | Appendix A, `DDD-measure-01` row | **defer** — generated | +| 1382 | Appendix A, `DDD-measure-11` row | **defer** — generated | +| 1420 | Appendix A, `term:verdict` row | **defer** — generated | + +**Measure note — 26 occurrences** + +*Move (16):* L27, L41, L114, L129, L466, L482, L493, L495, L601, L651, L713, L842, L899 — +thirteen straightforward — plus L22 and L88 (flagged, §2.4) and L464 (flagged, §2.5). + +*Defer (10):* + +| Line(s) | Why | +|---|---| +| 138, 484, 879 | each reproduces `DDD-measure-12`'s statement verbatim — *"fixed by the task, the tolerance, and the ground distribution"* — as an italicised quotation. The node still says `ground distribution` at `v5.12.0`. Renaming would make the note misquote a live claim | +| 102 | inside a blockquote restating `term:verdict`'s definition. The registry still reads `ground distribution` at `v5.12.0` | +| 923 | `measure-nonuniform-ground.py` — a **filename** for an asset in the principle repository's `core/assets/`. The audit's extract does not flag it as an identifier; this session does | +| 1001, 1010, 1011 ×2, 1035 | the note's own node table — canon's words | + +**Totals: 18 proposed moved, 13 deferred, of 31.** + +Reproducible: `w1-enumerate.py` in this directory prints the table and asserts that every S5 +occurrence the classification names in the two manuscripts carries exactly one disposition. It +fails loudly if the manuscript has moved under the enumeration — which it has, once, and the +next paragraph is that finding. + +**The classification's line numbers for Paper A are stale, and the staleness is instructive.** +The audit's extract was taken on 2026-08-24; the item-4 session regenerated Appendix A the next +day with a `kind` column, moving every appendix row down by eleven lines. So three of Paper A's +five S5 rows point at the wrong lines — `1365`, `1371`, `1409` in the extract against `1376`, +`1382`, `1420` at head. Nothing is lost, because the sense assignment travels with the +occurrence rather than the line, but a migration executed by line number would have edited three +wrong rows. The enumerator carries both numbers and reconciles neither silently. **The measure +note has not moved since the extract, and Paper A itself is unchanged since the review's +`40d277f` in every respect but that regeneration.** + +### 2.4 The ambiguous set, separated as the rule requires + +**The `ground the task faces` pair.** The classification splits a three-word difference across two +senses: *"the **distribution of** ground the task faces"* is S5 (Paper A L502; measure note L22, +L88), while *"over **the** ground the task faces"* is S1 (measure note L107, L911). Both readings +are defensible and the boundary is invisible to a reader. + +Renaming the S5 half and leaving the S1 half produces, inside one document, *"the deployment +distribution the task faces"* three lines from *"the ground the task faces"*. That is either the +migration doing exactly its job — disambiguating a word that carried two senses — or it is a +migration artefact that will read as an inconsistency to a reviewer who does not hold the sense +table. **This is a ruling, and it is Emil's.** The session's recommendation is to **move** the three +S5 members and leave the two S1 members untouched, because leaving them is what "renames the +population sense only" means; but the three are flagged rather than moved silently. + +**Two occurrences that are S5 by the audit and read as elliptical.** Measure note L482 (*"41% of +the whole on the benign ground, 57% on the uniform"*) and L493 (*"the skewed ground exercises it +with `P` varied"*) drop the noun the distribution qualifies. They are S5 — the three deployments +are the population — and they are listed as moves, but each needs a rewrite rather than a token +swap: *"on the benign deployment distribution"*, *"the skewed distribution"*. + +### 2.5 One coupling W1 creates, and it is not cosmetic + +Measure note §5.4's heading is *"`P` varied → non-uniform ground"*. Its reproducing asset is +`core/assets/measure-nonuniform-ground.py`, **upstream and out of this session's reach**. Renaming +the heading while the asset keeps its name puts a section and the script that computes it under two +different words. Three dispositions, and the session recommends the second: + +1. rename the heading and leave the asset — a visible divergence, honest, resolved by W2; +2. rename the heading and **say so in one clause** — *"worked by `measure-nonuniform-ground.py`, + named before the vocabulary moved"*; +3. defer the heading to W2, so section and asset move together. + +### 2.6 Two consequences of touching the measure note at all + +**Paper A pins the measure note by commit.** Its front matter cites the note *"by path at commit +`aa7e135`"*. The note is unchanged since `aa7e135` — verified. The moment W1 edits it, Paper A +either keeps citing a commit whose prose says `ground distribution` while Paper A's own says +`deployment distribution`, or advances the citation to a commit this session creates. **Ruling +needed at Gate 4;** the recommendation is to advance it, in the same commit, with the reason stated +in the front matter. + +**The measure note has no checkers.** Paper A has `check-quotations.py` and `check-appendix.py`; +the note has neither, and its node table is hand-carried at `v5.7.0`. So the ten deferrals in §2.3 +are protected by this session's reading and by nothing else. That is a finding about the note, and +it is why the deferral list is enumerated line by line above rather than summarised. + +--- + +## 3. R-1 — the survey plan + +**The survey is reading before it is drafting.** This section is the plan; the section itself is +Gate 2's, and no differentiation is claimed here that the reading has not yet earned. + +### 3.1 Locators can be verified in this session + +Confirmed: outbound HTTPS resolves and publisher pages, DOIs and repository records are reachable. +So the Hayek precedent's standard applies in full — **every locator verified directly, or flagged +unverified in its own entry with the reason.** The Tesler entry stays as the pattern for a source +with no primary publication. + +### 3.2 A first catch, before the survey starts + +The review's own locator for Hollnagel and Woods is +`10.1201/9781420005684`, given as *"Woods and Hollnagel, Joint Cognitive Systems"*. That DOI +resolves to **Woods & Hollnagel (2006), *Joint Cognitive Systems: Patterns in Cognitive Systems +Engineering***, CRC Press — **not** Hollnagel & Woods (2005), *Joint Cognitive Systems: Foundations +of Cognitive Systems Engineering*, CRC Press, ISBN 0-8493-2821-7, which is the volume that states +the joint cognitive system as a unit of analysis and is the one the review's prose is about. + +Two books, reversed author order, one year apart, sharing a main title. **This is the Hayek +facsimile check earning its keep on the first entry**, and it is the reason the plan below names an +edition for each work rather than a name. + +### 3.3 The seven entries, with candidate locators and the axis each is differentiated on + +Every axis below rests on **filed canon**. Q44 and Q45 sharpened the session's reading of where the +differences lie; they are unfiled, they are cited nowhere, and no distinction in the drafted section +will depend on them. Where a distinction needs them, it is written register-native as the paper's +own analysis or it waits. + +| Work | Candidate locator (to verify at Gate 2) | What the framework **takes** | Where it **differs**, and on which filed node | +|---|---|---|---| +| **Hutchins**, distributed cognition | *Cognition in the Wild*, MIT Press 1995, ISBN 9780262082310 (pb 9780262581462) | the arrangement, not the individual, as the cognitive unit — and the priority is Hutchins's, not the framework's | descriptive rather than governing: no assurance axis, no accountable principal. `term:arrangement`, `DDD-frame-03` | +| **Hollnagel & Woods**, joint cognitive systems | *Foundations*, CRC Press 2005, ISBN 0-8493-2821-7; and *Patterns*, Woods & Hollnagel 2006, doi:10.1201/9781420005684 | **the closest neighbour, and the one that most narrows the novelty claim** — the human–machine ensemble as one control process | the **supply partition**: where a resolution comes from versus what assures it, and standing versus occasioned supply. `DDD-frame-03`, `DDD-cost-09` | +| **Leveson**, STAMP | *Engineering a Safer World*, MIT Press 2011, ISBN 9780262016629; OA edition, ch. 4 | safety as a control-structure property, and inadequate control as the failure unit | **filing versus delivery**: STAMP's constraint can be authored and absent at the act, and the model has no place to record that. `DDD-delivery-01`, `DDD-delivery-02` | +| **Horvitz**, mixed-initiative | CHI '99, pp. 159–166, doi:10.1145/302979.303030 | initiative as an allocation decision per interaction, not a fixed assignment | **commitment levels** — outcome, policy, principal — as what the allocation is *of*. `term:commitment-level` | +| **Bovens**, accountability | *European Law Journal* 13(4):447–468, 2007, doi:10.1111/j.1468-0386.2007.00378.x | accountability as a **relation** to a forum, not a virtue of the actor — the framework's `DDD-frame-08` is the same move | the relation as a **design-time checkable property of an engineering arrangement**, with a named incompleteness. `DDD-frame-08`, `term:accountability` | +| **Matthias**, responsibility gap | *Ethics and Information Technology* 6(3):175–183, 2004, doi:10.1007/s10676-004-3422-1 | the gap as a real structural consequence of learning systems | the gap read as **an arrangement naming an executor and no principal** — fixable rather than novel. `DDD-frame-08` | +| **Meaningful human control** | Santoni de Sio & van den Hoven, *Frontiers in Robotics and AI* 5:15, 2018, doi:10.3389/frobt.2018.00015 | tracking and tracing as two separable conditions | **delivery**: a tracing condition satisfied in the record and failed at the act is *escape that presents as governance*. `DDD-delivery-02` | + +### 3.4 The standard each entry meets, and the shape of the section + +The measure note's §8 set the standard and it is the one to meet: **what the framework takes, and +where it differs** — not a citation list. Each entry closes on both, in that order, with the taking +first, because the review's finding is that the paper claimed absence it had not surveyed. + +Proposed shape, for ruling: the existing §11's ten short neighbourhood paragraphs are **kept and +compressed**, and the seven works above are added as a **comparison table plus one paragraph each** +— the reviewer's own recommendation in their §12.7. Estimated addition: **1,200–1,600 words**. + +### 3.5 What the survey may conclude, and what it may not + +The framework's genuine difference is more statable now than when the review was written. It is +still a difference the survey has to earn. **No entry will be drafted before its source is read**, +and where the reading does not support a difference, the entry will say the framework takes and +adds nothing — which is a finding, not a failure. + +--- + +## 4. R-2 — retitle candidates + +*The Missing Parameter* claims absence from prior work the paper has not surveyed. The triage +accepts the retitle. Four candidates, with what each costs. + +| # | Title | What it does | Cost | +|---|---|---|---| +| **1** | **Actor-Indexed Determination: where consequential choices are resolved, what assures them, and who answers** | makes **no absence claim at all**; promotes the paper's existing subtitle, which already describes the contribution accurately | loses the memorable phrase; the paper becomes eponymous with the repository | +| **2** | Indexing Determination to the Arrangement: a synthesis of resolution, assurance, delivery and accountability | states the **narrowed claim** in the title, in the reviewer's own terms | "synthesis" in a title is a hostage; it invites the question the survey must answer | +| **3** | A Missing Parameter in Complexity-Allocation Accounts: actor-indexed determination | closest to the reviewer's own proposal; **concedes visibly**, which has been this paper's credibility mechanism | keeps an absence claim, narrowed to a named literature the survey must then actually cover | +| **4** | Resolution, Assurance, Delivery, Accountability: an auditable framework for actor-indexed determination | leads with the four separated coordinates the reviewer called the durable core | reads as a list; buries the index | + +**Recommendation: 1, with 3 as the alternative.** Candidate 1 is the only one that cannot be +falsified by the survey it precedes. Candidate 3 keeps continuity with the merged version and +concedes in the title, which is a real virtue — but it makes a scoped absence claim, and the scope +("complexity-allocation accounts") is exactly the neighbourhood where Brooks, Tesler and Meyer sit +and where the claim is most nearly defensible. **Either is defensible; 2 and 4 are not +recommended.** + +--- + +## 5. R-4 — the supplement split's proposed boundary + +### 5.1 The boundary rule, stated so it can be applied rather than argued case by case + +> **A passage moves to the supplement when it is about the graph. It stays when it is about the +> world, or about how much warrant the paper has.** + +Filing history, minting decisions, pin mechanics, node tables and instrument descriptions are about +the graph. *"This claim is `projected`, its evidence field is empty, and the study is unrun"* is +about warrant, and it **stays** — the review praised the disclosure discipline twice and the triage +says explicitly not to trade it for brevity. + +### 5.2 What moves, and what it costs + +Into `papers/paper-a/paper-a-supplement.md`, with the generator retargeted at it: + +| Passage | Words | +|---|---| +| **Appendix A** in full — claims, decisions, terms, hypothesis-set tables | 3,542 | +| **Reproduction** — the three scripts and their invocation | 138 | +| The front matter's **pin apparatus** (one sentence stays in the body) | ~115 | +| **Note on claim status** — the five-value vocabulary list; a compressed honest paragraph stays | ~144 | +| §2.4's *"ruled ineligible for minting"* filing rationale | 52 | +| §3's *"both were minted only when a downstream projection needed them"* | 45 | +| §5.2's *"a dedicated node is pending filing on the open-questions wave"* | 116 | +| §11.1's *"it is not filed, and it is not filed for a stated reason"* | 70 | + +### 5.3 The arithmetic, stated before it is a surprise at Gate 4 + +Measured at head, by the same method the review used — regex word tokens over the Markdown source, +which reproduces the review's own figures to within 1%: + +| | Words | Body headings | +|---|---|---| +| Whole file | **15,250** (review: ~14,900) | — | +| Body, before Appendix A | **11,708** (review: ~11,600) | **63** (review: 63) | +| Appendix A | 3,542 | — | + +Removing everything in §5.2 takes the body to roughly **11,030**. **R-1 then adds 1,200–1,600.** + +**The supplement split does not by itself make the main paper shorter.** It lands at roughly +**12,200–12,600 words** — longer than it is now. Saying so at Gate 1 is cheaper than discovering it +at Gate 4. + +If a shorter main paper is the objective, the reduction has to come from argument, and the four +honest candidates are: §5.2's closure ladder (which is being repaired anyway — the triage records +the axis-mixing as Emil's own Gate 1 ruling, and the ladder should end at constructively-closed); +§8's worked example (six subsections, structural rather than numerical); §9.5's study design; and +§10.3's boundary cases. **This is a ruling and the session does not take it.** + +### 5.4 One thing the split must not do + +The apparatus **stays** — it moves. `check-quotations.py`, `check-appendix.py` and `gen-appendix.py` +keep running, against the supplement instead of the body, and the close report says so. A +supplement that the checkers no longer read would convert a verified apparatus into a claim about +one, which is the exact failure the review credited the paper for avoiding. + +--- + +## 6. What this gate did not do + +Nothing in `papers/` is touched. No claim, term or decision is filed, amended or retired in either +repository. The pin stays at `v5.9.0` until Gate 4, where the advance is made with the predicted +W6/W7 results **stated before the operation** and verified after. Q44, Q45 and the ontology +explainer are read and cited nowhere. diff --git a/meta/sessions/2026-08-30-paper-a-revision/status-sweep.py b/meta/sessions/2026-08-30-paper-a-revision/status-sweep.py new file mode 100755 index 0000000..71f7f77 --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/status-sweep.py @@ -0,0 +1,91 @@ +#!/usr/bin/env python3 +"""Inline status assertions in a projection, checked against the graph at a ref. + +The gap this closes. `check-quotations.py` verifies BLOCK quotations, and its docstring says so +honestly. But the manuscript's more common form is the inline citation that asserts a status in +prose -- `[DDD-measure-06, **established**]` -- and no instrument reads it. At the v5.9.0 -> v5.12.0 +advance that gap hides a retirement: DDD-measure-06 is retired at v5.12.0, the paper calls it +established three times, and both existing checkers pass those three lines. + +A status label is the part of a citation a reader converts into warrant. It is therefore the part +most worth checking, and it was the part with no check. + +Usage: status-sweep.py [] +Exit 0 when every inline status assertion matches the graph at ; 1 otherwise. +Appendix A is excluded: it is generated wholesale and re-read by check-appendix.py. +""" +import re +import subprocess +import sys + +import yaml + +STATUSES = r'(settled|established|projected|reported|draft|retired)' + + +def show(repo, ref, path): + r = subprocess.run(['git', '-C', repo, 'show', f'{ref}:{path}'], + capture_output=True, text=True) + if r.returncode: + sys.exit(f'cannot read {path} at {ref}: {r.stderr.strip()}') + return r.stdout + + +def statuses(repo, ref): + out = {} + for d in ('core/claims/', 'core/decisions/'): + listing = subprocess.run(['git', '-C', repo, 'ls-tree', '-r', '--name-only', ref, d], + capture_output=True, text=True).stdout.split() + for f in listing: + if f.endswith('.yaml'): + y = yaml.safe_load(show(repo, ref, f)) + out[y['id']] = y.get('status') + for t in yaml.safe_load(show(repo, ref, 'core/graph/terms.yaml'))['terms']: + out[t['id']] = t.get('status') + return out + + +def main(path, repo, ref, baseline=None): + now = statuses(repo, ref) + was = statuses(repo, baseline) if baseline else {} + body = open(path).read().split('## Appendix A')[0] + + checked = 0 + failures = [] + for i, line in enumerate(body.split('\n'), 1): + for m in re.finditer(r'\[([A-Za-z][A-Za-z:\-0-9]*)((?:,|\s\u2014)[^\]]*)\]', line): + node_id, tail = m.group(1), m.group(2) + s = re.search(r'\*{0,2}' + STATUSES + r'\*{0,2}', tail) + if not s: + continue + checked += 1 + claimed, actual = s.group(1), now.get(node_id) + if actual is None: + failures.append((i, node_id, claimed, 'ABSENT at ref')) + elif actual != claimed: + prior = f' (was {was[node_id]} at {baseline})' if node_id in was else '' + failures.append((i, node_id, claimed, f'{actual}{prior}')) + + # Nodes cited anywhere in the body whose status moved between the two refs, label or not: + # a status can move under a citation that never named one, and the prose around it still + # asserts the old reading. + moved = [] + if baseline: + cited = set(re.findall(r'DDD-[a-z]+-\d+', body)) | set(re.findall(r'term:[a-z-]+', body)) + moved = sorted((c, was.get(c), now.get(c)) for c in cited if was.get(c) != now.get(c)) + + print(f'{checked} inline status assertions checked, {len(failures)} wrong ' + f'({path} against {ref})') + for line, node_id, claimed, actual in failures: + print(f' L{line}: [{node_id}, {claimed}] -- graph says {actual}') + if baseline: + print(f'{len(moved)} cited node(s) whose status moved {baseline} -> {ref}:') + for c, o, n in moved: + print(f' {c}: {o} -> {n}') + return 1 if failures else 0 + + +if __name__ == '__main__': + if not 4 <= len(sys.argv) <= 5: + sys.exit(__doc__) + sys.exit(main(*sys.argv[1:])) diff --git a/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py b/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py new file mode 100644 index 0000000..0b6d69e --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py @@ -0,0 +1,103 @@ +#!/usr/bin/env python3 +"""W1's occurrences in the two manuscripts, enumerated with their dispositions. + +W1 renames the POPULATION sense of `ground` -- sense S5, "the population the task faces" -- to +`deployment distribution`. It touches no S1, S2 or S3 occurrence, and an occurrence ambiguous +between S5 and any other sense DEFERS to the migration rather than being ruled in a paper session. + +The classification is NOT re-derived here. It is the ground-migration session's, execution-grade +and committed, read through w0-full-v2.json against the ground audit's own extract. What this +script adds is the disposition: not every S5 occurrence is W1's to move, and the exclusions are +the substance. + +Four exclusions, each a reason a rename would assert something the session cannot: + + generated a row in a table `gen-appendix.py` regenerates from the graph. Renaming it is + hand-editing the appendix, which forges agreement the pin does not have. + quotes the passage reproduces a live claim's statement or a registry entry verbatim. Canon + still says `ground distribution` at v5.12.0; renaming would make the paper misquote. + identifier a filename. `measure-nonuniform-ground.py` is an asset in the principle repository + and out of this session's reach. The audit's extract does not flag it; this does. + ambiguous S5 by the classification and three words from an S1 reading of the same phrase. + +Usage: w1-enumerate.py [] +Prints the enumeration and the counts the close report must use. +""" +import collections +import json +import os +import sys + +TWO = ('papers/paper-a/paper-a.md', 'papers/measure-note/measure-note.md') + +# Dispositions, keyed by (path, line). Every line the classification calls S5 in the two +# manuscripts appears exactly once; the script asserts that, so a drifted manuscript fails +# loudly rather than silently dropping an occurrence. +# Lines are the AUDIT EXTRACT's, and for paper-a.md three of them are stale by design rather +# than by error: the extract was taken at the ground audit (2026-08-24), and the item-4 session +# regenerated Appendix A the next day with a `kind` column, moving every appendix row down. The +# head line is carried beside the extract line so a reader can find the occurrence in the file, +# and the divergence is recorded rather than silently reconciled -- an enumeration that quietly +# renumbers is an enumeration whose input cannot be checked. measure-note.md did not move. +DISPOSITION = { + ('papers/paper-a/paper-a.md', 502): ('move-flagged', 502, 'ambiguous: "distribution of ground the task faces" (S5) vs "the ground the task faces" (S1)'), + ('papers/paper-a/paper-a.md', 672): ('move', 672, ''), + ('papers/paper-a/paper-a.md', 1365): ('defer', 1376, 'generated: Appendix A, DDD-measure-01'), + ('papers/paper-a/paper-a.md', 1371): ('defer', 1382, 'generated: Appendix A, DDD-measure-11'), + ('papers/paper-a/paper-a.md', 1409): ('defer', 1420, 'generated: Appendix A, term:verdict'), +} +_MN = 'papers/measure-note/measure-note.md' +for ln in (27, 41, 114, 129, 466, 493, 495, 601, 651, 713, 842, 899): + DISPOSITION[(_MN, ln)] = ('move', ln, '') +DISPOSITION[(_MN, 482)] = ('move', 482, 'elliptical: rewrite, not a token swap') +DISPOSITION[(_MN, 22)] = ('move-flagged', 22, 'ambiguous: the "ground the task faces" pair') +DISPOSITION[(_MN, 88)] = ('move-flagged', 88, 'ambiguous: the "ground the task faces" pair') +DISPOSITION[(_MN, 464)] = ('move-flagged', 464, 'heading couples to upstream asset measure-nonuniform-ground.py') +for ln in (138, 484, 879): + DISPOSITION[(_MN, ln)] = ('defer', ln, 'quotes: DDD-measure-12 statement, verbatim') +DISPOSITION[(_MN, 102)] = ('defer', 102, 'quotes: term:verdict definition, blockquoted') +DISPOSITION[(_MN, 923)] = ('defer', 923, 'identifier: measure-nonuniform-ground.py') +for ln in (1001, 1010, 1011, 1035): + DISPOSITION[(_MN, ln)] = ('defer', ln, "quotes: the note's own node table") + + +def main(root='.'): + ses = os.path.join(root, 'meta', 'sessions') + rows = json.load(open(os.path.join(ses, '2026-08-24-ground-audit', 'classification.json'))) + sense = json.load(open(os.path.join(ses, '2026-08-27-ground-migration', 'w0-full-v2.json'))) + for i, r in enumerate(rows): + r['s'] = sense[str(i)] + + s5 = [r for r in rows if r['s'] == 'S5'] + mutable = [r for r in s5 if not r['immutable']] + print(f'S5 corpus-wide: {len(s5)} immutable {len(s5) - len(mutable)} ' + f'MUTABLE {len(mutable)} mutable excl. product-cli ' + f'{sum(1 for r in mutable if r["repo"] != "product-cli")}') + + here = [r for r in mutable if r['repo'] == 'downstream' and r['path'] in TWO] + print(f'in the two manuscripts: {len(here)} ' + f'{dict(collections.Counter(r["path"] for r in here))}') + print(f'anywhere under papers/: ' + f'{sum(1 for r in mutable if r["repo"] == "downstream" and r["path"].startswith("papers/"))}') + + # L1011 carries two occurrences on one line; count rows, key dispositions by line. + counts = collections.Counter() + print() + for r in sorted(here, key=lambda r: (r['path'], r['line'])): + key = (r['path'], r['line']) + if key not in DISPOSITION: + sys.exit(f'UNRULED S5 occurrence {key} -- the manuscript moved under the enumeration') + verdict, head_line, why = DISPOSITION[key] + counts[verdict.split('-')[0]] += 1 + moved_line = '' if head_line == r['line'] else f' (head L{head_line})' + print(f' {verdict:13s} {r["path"].split("/")[-1]:16s} ' + f'L{r["line"]:<5d}{moved_line:<12s} {why}') + + print(f'\nMOVED {counts["move"]} DEFERRED {counts["defer"]} ' + f'of {sum(counts.values())} in the two manuscripts, of {len(mutable)} corpus-wide') + print('The corpus-wide mutable total is 88. The two manuscripts hold ' + f'{len(here)} of it, which is {100 * len(here) // len(mutable)}% -- not 88%.') + + +if __name__ == '__main__': + main(*sys.argv[1:]) From 689c62413f48d1e44835f4e8111f99059746eafb Mon Sep 17 00:00:00 2001 From: Claude Date: Sun, 30 Aug 2026 19:28:24 +0000 Subject: [PATCH 3/9] GATE 2 (draft-pending-ruling): the related-work survey, read then drafted MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit R-1 drafted. §11 is rewritten: seven inherited neighbourhoods compressed and kept, the seven works the review named added with what the framework takes and where it differs, a comparison table, and the narrowed novelty claim. The old §11.1 becomes §11.6 unchanged. READING BEFORE DRAFTING. gate2-survey-notes.md is the reading record: how each locator was verified (Crossref, Open Library, NCBI — no publisher page was reachable and none is the basis of any entry), and at what grade each text was read. Four works were read primary and in full; the two Hollnagel and Woods volumes were not obtained, and both the notes and the reference entries say so. Verifying a locator is not reading a text, and the References preamble now marks the two separately. TWO CATCHES. The review's own locator for Hollnagel and Woods resolves to Woods & Hollnagel (2006), PATTERNS, while its prose concerns Hollnagel & Woods (2005), FOUNDATIONS — two books, reversed author order, one shared main title. Both are cited; the error goes to the response-to-review rather than being fixed quietly, on Emil's ruling. And a "law of conservation of complexity" attributed by a search result to Patterns could not be corroborated and is used nowhere; it is written down so a later session does not rediscover and believe it. THE SURVEY CORRECTED A DIFFERENCE CLAIM. The Gate 1 plan said Matthias's responsibility gap reads as "an arrangement naming an executor and no principal — fixable rather than novel". The reading does not support "fixable": the gap rests on a control condition on just ascription, and whether structural completeness answers it is a normative question DDD-frame-08 is projected about. The claim is withdrawn in the section itself rather than carried quietly. Three entries now record where a neighbour is stronger than the framework, Horvitz's decision criterion first among them. The section adds no node to Appendix A and no pin: all three checkers are green at the pin — 29 verbatim / 0 failing, 72 nodes / 0 discrepancies, 75 inline status assertions / 0 wrong. ALSO LANDED, on Gate 1 rulings: the stale-line-number finding added to the migration seed's method rule ("classification data cites content, never position"), with W1's counts corrected there from 29 to 31 in the two manuscripts and the corpus-wide 88 restated; and the ambiguous set's five occurrences moved to defer in w1-enumerate.py, which now reports 15 moved, 16 deferred. LENGTH, honestly. Body 11,708 -> 14,069 words; total 15,250 -> 17,611; headings 63 -> 68. That is over the 1,200-1,600 estimated at Gate 1, and the overrun is qualification: the per-entry reading grade, the comparison table, and the three places the framework comes off worse all cost words. Basis: DDD-frame-01; DDD-frame-03; DDD-frame-05; DDD-frame-08; DDD-frame-13; DDD-cost-09; DDD-delivery-01; DDD-delivery-02; DDD-measure-01; term:arrangement; term:accountability; term:commitment-level; DDD-dec-20 --- meta/migration-plan-ground.md | 36 ++- .../gate2-survey-notes.md | 160 +++++++++++ .../w1-enumerate.py | 12 +- papers/paper-a/paper-a.md | 260 ++++++++++++++++-- 4 files changed, 436 insertions(+), 32 deletions(-) create mode 100644 meta/sessions/2026-08-30-paper-a-revision/gate2-survey-notes.md diff --git a/meta/migration-plan-ground.md b/meta/migration-plan-ground.md index 714ec1b..a774595 100644 --- a/meta/migration-plan-ground.md +++ b/meta/migration-plan-ground.md @@ -29,12 +29,22 @@ projection: **380 (67%) need no edit; 187 are bare and need re-expression.** | Wave | Content | Size | Pins | |---|---|---|---| -| **W1** | S5 → *deployment distribution* | 87 mutable, 29 in merged papers | `term:verdict` unpinned | +| **W1** | S5 → *deployment distribution* | 87 mutable, **31** in the two manuscripts (was recorded as 29) | `term:verdict` unpinned | | **W2** | 11 registry entries + 11 embeds + upstream `core/` bare prose + **`apparatus/`** | 11 + 11 + 65 | **7 fire W6** | | **W3** | projections, applications, papers, downstream `core/` | 122 bare | 0 | | **W4** | `product-cli` | 1,203 | ruled separately | | **never** | U (261, of which **142 identifiers**) + immutable (260) | 521 | — | +**W1's counts, corrected again by the Paper A revision (2026-08-30).** Against +`w0-full-v2.json`: S5 totals **97**, of which 9 are immutable, leaving **88 mutable** corpus-wide +and **87** excluding `product-cli`. The two merged manuscripts hold **31** of those (measure note +26, Paper A 5) — **35%, not 88%**, and the row above is corrected from 29. Of those 31, Emil's Gate 1 +rulings put **15 in scope for that session and 16 out of it**: generated or hand-carried node-table +rows, passages quoting a live claim verbatim while canon still says `ground distribution`, one +filename, and the three occurrences ruled ambiguous between S5 and S1. **On those rulings W1's +remaining surface is 73 of 88, not zero** — the executed figure is recorded at that session's close +and supersedes this line if it differs. + **Corrected counts, after W0 and W0-bis:** S2 1,529 · S1 **567** · S3 369 · U 261 · S5 97 · S4 **22**. Canon is **S1-led** (182 S1 : 154 S3) and **canon's S4 is mechanically zero.** @@ -106,3 +116,27 @@ Three mechanisms, all found by reading rows the instrument had already scored: per-rule precision, never one figure for a table. A windowed rule earns a sample of its own.** **This closes two mechanisms, not the possibility of a rare third.** + +### A third, found by the Paper A revision and added on Emil's ruling + +> **Classification data cites content, never position.** + +The W0 classification's line numbers for `papers/paper-a/paper-a.md` are stale, and not by error: +the audit's extract was taken on 2026-08-24, and the item-4 session regenerated that paper's +Appendix A the next day with a `kind` column, moving every appendix row down by eleven lines. Three +of the five S5 rows in that file therefore point at the wrong lines — `1365`, `1371`, `1409` in the +extract against `1376`, `1382`, `1420` at head. + +Nothing was lost, because a sense assignment travels with the occurrence rather than with the line. +But **a migration executed by line number would have edited three wrong rows**, and the three wrong +rows were all inside a generated appendix, where an edit forges agreement with the graph. The +failure would have been silent in the only place the corpus cannot afford silence. + +The rule is cheap: an occurrence is addressed by its file plus a content anchor, and a line number +is a convenience for a reader rather than an instruction to an instrument. `w1-enumerate.py` +(`meta/sessions/2026-08-30-paper-a-revision/`) carries both numbers and reconciles neither silently +— it fails loudly when a manuscript has moved under the enumeration, which is how this was found. + +**Corollary, and it is the wave-planning fact:** a generated table's rows are not the wave's to +move at all. They carry canon's words and they move when canon moves. W1's own scope shrinks once +that is applied — see the counts correction below. diff --git a/meta/sessions/2026-08-30-paper-a-revision/gate2-survey-notes.md b/meta/sessions/2026-08-30-paper-a-revision/gate2-survey-notes.md new file mode 100644 index 0000000..afb198b --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/gate2-survey-notes.md @@ -0,0 +1,160 @@ +# GATE 2 — the survey's reading record + +**Status: draft-pending-ruling.** The record of what was read, how each locator was verified, and +what the reading did and did not support. Filed separately from the drafted section because the +survey's warrant is the reading, and a reader who wants to check the section should not have to +reconstruct the reading from it. + +**The standard, from Emil's Gate 1 ruling and the Hayek precedent:** every locator verified +directly, or flagged unverified in its own entry with the reason. The Tesler entry is the pattern +for a source with no primary publication. + +--- + +## 1. How each locator was verified + +Two registries were used, both authoritative and both independent of the review: **Crossref**'s +DOI records (`api.crossref.org`) and **Open Library**'s ISBN records. Publisher pages +(`mitpress.mit.edu`, `direct.mit.edu`, `taylorfrancis.com`, `routledge.com`, `dl.acm.org`) refused +automated requests, so none of them is the basis of any entry. + +| Work | Locator, as verified | Verified against | +|---|---|---| +| Hutchins | *Cognition in the Wild*, MIT Press 1995, ISBN 9780262082310; pb Bradford Books 1996, ISBN 9780262581462 | Open Library, both ISBNs | +| Hollnagel & Woods | *Joint Cognitive Systems: Foundations of Cognitive Systems Engineering*, CRC Press 2005, ISBN 0-8493-2821-7, doi:10.1201/9781420038194 | Crossref **and** Open Library | +| Woods & Hollnagel | *Joint Cognitive Systems: Patterns in Cognitive Systems Engineering*, CRC Press 2006, ISBN 0-8493-3933-2, doi:10.1201/9781420005684 | Crossref **and** Open Library | +| Hollnagel & Woods (1983) | *International Journal of Man-Machine Studies* 18(6):583–600; reprinted *IJHCS* 51(2):339–356, 1999, doi:10.1006/ijhc.1982.0313 | NCBI E-utilities, PMID 11543350 | +| Leveson | *Engineering a Safer World*, MIT Press 2011, ISBN 9780262016629; pb 2017, ISBN 9780262533690; open-access edition | Open Library, both ISBNs | +| Horvitz | CHI '99, pp. 159–166, doi:10.1145/302979.303030 | Crossref, **and the full text read** | +| Bovens | *European Law Journal* 13(4):447–468, 2007, doi:10.1111/j.1468-0386.2007.00378.x | Crossref, **and the full text read** | +| Matthias | *Ethics and Information Technology* 6(3):175–183, 2004, doi:10.1007/s10676-004-3422-1 | Crossref, **and the full text read** | +| Santoni de Sio & van den Hoven | *Frontiers in Robotics and AI* 5:15, 2018, doi:10.3389/frobt.2018.00015 | Crossref | + +--- + +## 2. Two catches, and they are the reason the rule exists + +### 2.1 The review's own Hollnagel-and-Woods locator is the wrong volume + +The review cites `10.1201/9781420005684` and names it *"Woods and Hollnagel, Joint Cognitive +Systems"*. Crossref resolves that DOI to: + +> **Woods, David D.; Hollnagel, Erik. *Joint Cognitive Systems: Patterns in Cognitive Systems +> Engineering.* CRC Press, 27 March 2006.** + +The volume the review's **prose** is about — the joint cognitive system stated as a unit of +analysis — is the sibling: + +> **Hollnagel, Erik; Woods, David D. *Joint Cognitive Systems: Foundations of Cognitive Systems +> Engineering.* CRC Press, 28 February 2005.** doi:10.1201/9781420038194. + +Two books, one year apart, sharing a main title, with the author order reversed between them. A +citation database would have propagated the error silently, which is exactly what the Hayek +facsimile check was instituted to prevent. **Both volumes are cited in the drafted section**, since +both are on point. Per Emil's Gate 1 ruling, the error is recorded for the response-to-review +rather than quietly corrected: telling the reviewer their own citation was reversed is more useful +to them than fixing it behind their back. + +### 2.2 A conservation law attributed to *Patterns* — could not be corroborated, and is not used + +A search result attributed a **"Law of Conservation of Complexity"** — *"complexity is conserved +under transformation and translation"* — to Woods & Hollnagel's *Patterns*. If true it would be +directly load-bearing: it would give the framework's own conservation principle a named precedent +inside the literature the review says is unengaged, and it would change the Tesler entry as well. + +**It could not be corroborated.** A curated index of the resilience-engineering literature's named +laws attributes the *law of fluency*, the *law of stretched systems*, the *law of coordinative +entropy* and the *law of the kludge* to Hoffman & Woods's *Beyond Simon's Slice*, and records no +conservation law from *Patterns* at all. No primary text was obtained. + +**Nothing in the drafted section rests on it.** It is written down here so that a later session does +not rediscover the same unverified attribution and believe it the second time. If someone with the +book in hand can confirm or refute it, that is a real finding either way, and it is booked as a +successor item rather than a footnote. + +--- + +## 3. What was read, and at what grade + +The distinction matters and the section is written to respect it. **Locator verified** is not +**text read**, and a difference claim is only as good as the reading behind it. + +| Work | Grade of reading | +|---|---| +| Horvitz 1999 | **primary, in full** — all twelve critical factors read verbatim from the author's own copy of the paper | +| Bovens 2007 | **primary, in full** — the definition, the three stages, the forum typology and the problem of many hands read verbatim | +| Matthias 2004 | **primary, in full** — the gap's statement and the control condition it rests on read verbatim | +| Santoni de Sio & van den Hoven 2018 | **primary** — the tracking and tracing conditions read in the published full text | +| Hollnagel & Woods 1983 | **primary abstract** — the definition of a cognitive system, from the authors' own abstract via NCBI's record | +| Hutchins 1995 | **author's own summary plus secondary** — the unit-of-analysis and emergent-properties formulations from the author's page for the book | +| Hollnagel & Woods 2005 / Woods & Hollnagel 2006 | **secondary only** — the books themselves were not obtained | +| Leveson 2011 | **secondary, but authored by Leveson** — the STPA Handbook's statement of the control structure, process models and the four types of unsafe control action, read in full | + +**The two weakest readings are on the closest neighbour**, which is uncomfortable and is stated +rather than hidden. The Hollnagel-and-Woods entry is therefore written to claim only what the +primary 1983 abstract and multiple independent secondary characterisations agree on, and it credits +the priority rather than arguing a difference the reading has not earned. + +--- + +## 4. Substance, per work, with the differentiating node + +Every difference below cites a node **already filed and already cited elsewhere in the manuscript**, +so the drafted section adds no node to Appendix A and no pin to `graph/upstream.yaml`. Q44 and Q45 +sharpened the reading; nothing here depends on them and neither is cited. + +**Hutchins.** Takes: the unit of cognitive analysis is a collection of people and artefacts in a +work practice, and *"groups must have cognitive properties that are not predictable from a knowledge +of the properties of the individuals in the group"*. Differs: descriptive, not governing — no +acceptance relation, no assurance mechanism, no accountable principal. `term:arrangement`, +`DDD-frame-03`. + +**Hollnagel & Woods.** Takes: **the priority for the arrangement as the unit is theirs, not the +framework's.** CSE *"introduces the concept of a cognitive system: an adaptive system which +functions using knowledge about itself and the environment in the planning and modification of +actions"* (1983 abstract, verbatim), and the joint cognitive system treats people and machines as +one such system. Differs: the JCS is characterised by what it does; it does not separate where a +resolution comes from from what establishes its acceptability, nor standing supply from occasioned. +`DDD-frame-03`, `DDD-cost-09`. + +**Leveson.** Takes: a hierarchical control structure whose vertical axis *"indicates control and +authority"*, controllers with process models, and four provably complete types of unsafe control +action — not providing, providing, wrong timing or order, wrong duration. Differs: the four types +quantify over control actions **provided**, and a constraint that is authored and absent at the act +is modelled as a process-model or control-algorithm flaw. STPA's causal-scenario step can reach the +case; what is absent is a standing category that separates a decision's being **filed** from its +being **delivered at an act-site**. `DDD-delivery-01`, `DDD-delivery-02`. + +**Horvitz.** Takes: initiative as an allocation decided per interaction under uncertainty about the +user's goal, with the twelve factors including *"inferring ideal action in light of costs, benefits, +and uncertainties"* and *"minimizing the cost of poor guesses about action and timing"*. Differs: +the factors govern **when the automated service should act**, not what the arrangement has committed +in advance; commitment levels are what the allocation is of. `term:commitment-level`. **And it +differs in the framework's disfavour**: Horvitz supplies a decision-theoretic criterion for the +allocation, and the framework supplies none outside the closing region. That is said in the entry. + +**Bovens.** Takes: accountability as *"a relationship between an actor and a forum"*, with three +stages — the actor obliged to inform, the forum able to interrogate (*"hence the close semantic +connection between 'accountability' and 'answerability'"*), and the forum able to pass judgement and +impose sanctions. Also the **problem of many hands**, which is §7's difficulty stated twenty years +earlier. Differs: **four of `DDD-frame-08`'s five elements are already Bovens's** — the entry says +so. The fifth, authority linkage, is not; and Bovens's relation is assessed retrospectively over +conduct where the framework's is a design-time property of an engineering arrangement. +`DDD-frame-08`, `term:accountability`. + +**Matthias.** Takes: the gap arises because *"nobody has enough control over the machine's actions +to be able to assume the responsibility for them"* — a control condition on just ascription, not a +missing name. Differs, **and less than the Gate 1 plan claimed**: the plan said the gap reads as +*"an arrangement naming an executor and no principal — fixable rather than novel"*. The reading does +not support *fixable*. `DDD-frame-08` holds that accountability does not read the executor's kind, +so an arrangement can be made complete without its principal predicting the executor; **whether that +completeness is morally sufficient to answer Matthias is a normative question the claim is +`projected` about and the paper does not settle.** This correction is the survey doing its job. + +**Meaningful human control.** Takes: two separable conditions — **tracking**, that the system *"be +responsive to the human moral reasons relevant in the circumstances"*, and **tracing**, that its +actions *"be traceable to a proper moral understanding on the part of one or more relevant human +persons"*. Differs: tracing is satisfied by a person who understands and is answerable; the +framework's delivery question is whether the governing decision **reached the act**, and a tracing +condition satisfied in the record while failing at the act is escape that presents as governance. +`DDD-delivery-02`. The two are adjacent and not the same, and the entry says which is which. diff --git a/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py b/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py index 0b6d69e..073ecc4 100644 --- a/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py +++ b/meta/sessions/2026-08-30-paper-a-revision/w1-enumerate.py @@ -19,6 +19,10 @@ identifier a filename. `measure-nonuniform-ground.py` is an asset in the principle repository and out of this session's reach. The audit's extract does not flag it; this does. ambiguous S5 by the classification and three words from an S1 reading of the same phrase. + RULED AT GATE 1 (Emil): all five members of the "ground the task faces" set defer, + including the three that read as S5. "Three words apart is exactly where a rule fails, + and W1's whole licence is that it renames a sense touching no other. Where that is + contestable, it isn't W1." Usage: w1-enumerate.py [] Prints the enumeration and the counts the close report must use. @@ -40,7 +44,7 @@ # and the divergence is recorded rather than silently reconciled -- an enumeration that quietly # renumbers is an enumeration whose input cannot be checked. measure-note.md did not move. DISPOSITION = { - ('papers/paper-a/paper-a.md', 502): ('move-flagged', 502, 'ambiguous: "distribution of ground the task faces" (S5) vs "the ground the task faces" (S1)'), + ('papers/paper-a/paper-a.md', 502): ('defer', 502, 'ambiguous, RULED defer: "distribution of ground the task faces" (S5) vs "the ground the task faces" (S1)'), ('papers/paper-a/paper-a.md', 672): ('move', 672, ''), ('papers/paper-a/paper-a.md', 1365): ('defer', 1376, 'generated: Appendix A, DDD-measure-01'), ('papers/paper-a/paper-a.md', 1371): ('defer', 1382, 'generated: Appendix A, DDD-measure-11'), @@ -50,9 +54,9 @@ for ln in (27, 41, 114, 129, 466, 493, 495, 601, 651, 713, 842, 899): DISPOSITION[(_MN, ln)] = ('move', ln, '') DISPOSITION[(_MN, 482)] = ('move', 482, 'elliptical: rewrite, not a token swap') -DISPOSITION[(_MN, 22)] = ('move-flagged', 22, 'ambiguous: the "ground the task faces" pair') -DISPOSITION[(_MN, 88)] = ('move-flagged', 88, 'ambiguous: the "ground the task faces" pair') -DISPOSITION[(_MN, 464)] = ('move-flagged', 464, 'heading couples to upstream asset measure-nonuniform-ground.py') +DISPOSITION[(_MN, 22)] = ('defer', 22, 'ambiguous, RULED defer: the "ground the task faces" set') +DISPOSITION[(_MN, 88)] = ('defer', 88, 'ambiguous, RULED defer: the "ground the task faces" set') +DISPOSITION[(_MN, 464)] = ('move-flagged', 464, 'UNRULED at Gate 1: heading couples to upstream asset measure-nonuniform-ground.py; re-raised at Gate 4') for ln in (138, 484, 879): DISPOSITION[(_MN, ln)] = ('defer', ln, 'quotes: DDD-measure-12 statement, verbatim') DISPOSITION[(_MN, 102)] = ('defer', 102, 'quotes: term:verdict definition, blockquoted') diff --git a/papers/paper-a/paper-a.md b/papers/paper-a/paper-a.md index d31097f..1d07de7 100644 --- a/papers/paper-a/paper-a.md +++ b/papers/paper-a/paper-a.md @@ -1113,54 +1113,205 @@ become decisions with no link back to the declared claims — escape by delivery ## 11. Related work -Each neighbourhood below is taken in turn, and each entry closes on what this paper takes from it -and what it does not. +**DRAFT — pending ruling.** This section is rewritten at the external review's §8, which found the +paper's novelty claim unestablished against literatures it had not surveyed. The finding was +correct. Two questions are kept apart throughout: what the framework **inherits**, which is most of +what follows, and what it **adds**, which is narrower than the merged version implied. + +Locators for every work named here are verified in the references, or flagged unverified there with +the reason. + +### 11.1 What the framework inherits **The determinable literature.** The determinable/determinate relation is W. E. Johnson's (*Logic* Part I, ch. XI, 1921), developed by Prior (*Mind* 58, 1949, in two parts) and Funkhouser (*The Logical Structure of Kinds*, OUP, 2014), and surveyed by Wilson (*SEP*, rev. 2023). Four structural gifts are taken directly: that the relation is not genus-and-species; that determinates under one determinable are constitutively exclusive; that determinateness comes in orders; and that the -determinable is a dimension of comparability. **What is the framework's, and not theirs**, is the -identification of the determinable as determination demand's object and of the verdict variable's -support as the determinate-space at the declared grain [DDD-frame-13, credits]. +determinable is a dimension of comparability. What is the framework's is the identification of the +determinable as determination demand's object and of the verdict variable's support as the +determinate-space at the declared grain [DDD-frame-13, projected — a proposal with a declared +falsifier, not a result]. **Cybernetics and requisite variety.** Ashby supplies the move this paper generalises — describing a regulator by what it must absorb rather than by what it is. What the framework adds is that the -absorbing arrangement is itself a variable, and that its coordinates can be enumerated (§1). +absorbing arrangement is itself a variable, and that its coordinates can be enumerated (§1). That +enumeration is `projected` [DDD-frame-01]. **Complexity allocation.** Brooks and Tesler ask where complexity resides and where it is moved. -The framework indexes both questions (§6.3) and, on the closing region, measures the second one +The framework indexes both questions (§6.3) and, on the closing region only, measures the second (§4.4). It does not claim their results as instances of its own. **Formal verification and proof-carrying systems.** The producer-independence result (§5.4) is the -framework's statement of what a sound check buys, deliberately narrowed to the checked property. -The literature's own scope conditions — soundness, ground completeness, checker legitimacy — are -carried inside the claim rather than assumed away. +framework's statement of what a sound check buys, deliberately narrowed to the checked property +[DDD-frame-05, projected]. The literature's own scope conditions — soundness, ground completeness, +checker legitimacy — are carried inside the claim rather than assumed away. **Design by contract.** Meyer's preconditions, postconditions and invariants are outcome-level and policy-level commitments written at an interface (§3.1). The framework contributes the level -vocabulary and the observation that contracts do not reach principal-level commitment at all. - -**Mixed-initiative and human-in-the-loop systems.** This literature studies arrangements directly -and is the framework's closest empirical neighbour. What the framework adds is a vocabulary for -*why* a given mix works — which commitments attach where, which predicates close, who answers — -rather than a catalogue of mixes that do. +vocabulary [term:commitment-level, draft] and the observation that contracts do not reach +principal-level commitment at all. **Bounded rationality and principal–agent theory.** The accountability relation (§7) is adjacent to principal–agent analysis and is not reducible to it: authority linkage and sanction path are institutional facts, not incentive parameters, and an arrangement can be incentive-aligned and -accountability-incomplete. - -**Algorithmic accountability and responsibility gaps.** The framework's contribution here is -[DDD-frame-08]'s structural reading: a responsibility gap is an arrangement naming an executor and -no principal, which is a fixable property of the arrangement rather than a novel moral category. +accountability-incomplete. That reading is `projected` [DDD-frame-08]. **Information theory.** Shannon's entropy and chain rule are used exactly as stated, and the theorem is Shannon's, always. The identification of determination demand with verdict entropy is -the framework's modelling claim, and it is the falsifiable part (§4.4) [DDD-measure-01]. - -### 11.1 A force the framework names and has not filed +the framework's modelling claim, and it is the falsifiable part (§4.4) [DDD-measure-01, projected]. + +### 11.2 The arrangement as the unit of analysis is not this framework's + +This is the review's central correction and the paper accepts it without qualification. **The +proposition that consequential work should be analysed at the level of an arrangement rather than an +individual actor is decades old, and the priority belongs to others.** + +**Hutchins, distributed cognition.** *Cognition in the Wild* takes as its unit of cognitive analysis +a collection of people and artefacts in a work practice, and argues that groups have cognitive +properties not predictable from the properties of their members. The framework's [term:arrangement] +is that move, applied to determination rather than to cognition. **(analysis)** What the framework +does not find in Hutchins is a governing apparatus: distributed cognition describes how a system +computes, and does not ask what acceptance relation the result must satisfy, what assures it, or who +answers for it. The addition is the governing questions, not the unit. + +**Hollnagel and Woods, joint cognitive systems.** This is the closest neighbour and the one that +most narrows what this paper may claim. Cognitive systems engineering *"introduces the concept of a +cognitive system: an adaptive system which functions using knowledge about itself and the +environment in the planning and modification of actions"* (Hollnagel & Woods 1983), and the joint +cognitive system reads people and machines together as one such system rather than as an interacting +pair. §2.5's insistence that the unit of comparison is the arrangement and not the isolated actor is +**their point before it is this framework's**, and the paper states that rather than rediscovering +it. + +**(analysis)** Where the framework goes further is not the unit but a partition inside it. A joint +cognitive system is characterised by what it does. This framework separates *where a resolution +comes from* from *what establishes its acceptability* [DDD-frame-03, projected], and then separates +assurance supplied **standing** — in a mechanism, independent of the act — from assurance supplied +**occasioned**, at the act and at the carrier's price [DDD-cost-09, projected]. Both are `projected`: +they are proposals with declared falsifiers, and neither has been tested against a joint cognitive +system that resists the partition. The claim here is that the distinction is available and useful, +not that it is established. + +**Leveson, STAMP and systems safety.** *Engineering a Safer World* models safety as a control +problem: a hierarchical control structure whose vertical axis carries control and authority, +controllers holding process models of what they control, and safety enforced as constraints on +behaviour. STPA classifies unsafe control actions in four ways — not provided, provided, provided +too early or too late or out of order, and stopped too soon or lasting too long — and the +classification is argued to be complete over control actions. The framework's accountability +material is weaker than Leveson's on authority, which STAMP carries structurally, and this paper +does not improve on it. + +**(analysis)** The distinction the framework adds is **filing versus delivery**. STPA's four types +quantify over control actions the controller *provides*; a constraint that was authored, is correct, +and is simply not present to the controller at the act appears as a flawed process model or a flawed +control algorithm. STPA's causal-scenario step can reach that case. What the framework supplies is a +standing category rather than a scenario: delivery is a property of a decision **at an act-site**, +the same decision can be delivered at one act-site and not at another, and governance filed but not +delivered is escape whose distinguishing feature is that the ledger shows coverage [DDD-delivery-01, +DDD-delivery-02, both projected]. Whether that category earns its place against STPA's scenarios is +an empirical question neither literature has settled. + +### 11.3 Allocating work inside the arrangement + +**Horvitz, mixed-initiative interaction.** *Principles of Mixed-Initiative User Interfaces* treats +initiative as something allocated per interaction under uncertainty about the user's goal, and gives +twelve factors for doing it — among them inferring ideal action in light of costs, benefits and +uncertainties; employing dialogue to resolve key uncertainties; and minimising the cost of poor +guesses about action and timing. This is the framework's closest neighbour on allocation, and it is +prior. + +**(analysis)** Two differences, and the first runs against the framework. **Horvitz supplies a +decision-theoretic criterion for the allocation and this framework supplies none** outside the +region where the acceptance predicate closes; off that region the framework offers vocabulary and a +governance question, not a rule for deciding. That is a real advantage of the mixed-initiative +account and the paper does not dispute it. The second difference is the framework's: Horvitz's +factors govern **when the automated service should act**, and are silent on what the arrangement +has fixed in advance. The level vocabulary names that — outcome, policy or principal +[term:commitment-level, draft] — and the observation the framework offers is that a mixed-initiative +system is one that has committed at policy level while leaving the act's initiative open. + +### 11.4 Accountability, and the gap + +**Bovens, the actor–forum relation.** *Analysing and Assessing Accountability* defines accountability +as *"a relationship between an actor and a forum, in which the actor has an obligation to explain and +to justify his or her conduct, the forum can pose questions and pass judgement, and the actor may +face consequences"*, in three stages: the actor is obliged to inform; the forum can interrogate — +*"hence the close semantic connection between 'accountability' and 'answerability'"* — and the forum +may pass judgement and impose sanctions. + +**The correspondence with [DDD-frame-08] is close, and the paper states how close.** **(analysis)** +Four of the framework's five elements are already Bovens's: a persistent principal is his actor, the +obligation to inform is attribution and answerability, and judgement with sanction is his third stage. +The element that is not his is **authority linkage** — that the acting party was entitled to act — +which his relation does not require because it assesses conduct after the fact rather than +constituting an arrangement before it. That is where the framework's contribution sits, and it is +one element and a change of tense, not a new theory of accountability. Bovens's **problem of many +hands** — that a forum confronted with an outcome often cannot unravel who contributed what — is §7's +difficulty stated in 2007, and the framework's answer to it is structural rather than forensic: the +arrangement is the unit, and the principal is named at its boundary [term:arrangement, +term:accountability]. + +**Matthias, the responsibility gap.** *The responsibility gap* argues that for learning machines the +manufacturer or operator cannot in principle predict future behaviour, so *"nobody has enough control +over the machine's actions to be able to assume the responsibility for them"*. The gap is grounded in +a **control condition on just ascription**, not in a missing name. + +**(analysis)** That matters for what this framework may say about it. [DDD-frame-08] holds that +accountability is a relation and not an intrinsic capacity, so the relation does not read the +executor's kind — which means an arrangement can be made accountability-complete without its +principal being able to predict the executor. **It does not follow that the gap is thereby closed.** +Whether structural completeness is morally sufficient where the control condition fails is a +normative question, [DDD-frame-08] is `projected` about exactly that, and this paper does not settle +it. An earlier draft of this section claimed the gap was "fixable rather than novel"; the reading +does not support that, and the claim is withdrawn here rather than carried quietly. + +**Meaningful human control.** Santoni de Sio and van den Hoven propose two conditions: **tracking**, +that a decision-making system *"should demonstrably and verifiably be responsive to the human moral +reasons relevant in the circumstances"*, and **tracing**, that its actions *"should be traceable to a +proper moral understanding on the part of one or more relevant human persons who design or interact +with the system"*. The separation of the two is the contribution, and it is theirs. + +**(analysis)** Tracing and the framework's delivery question are adjacent and are not the same, and +the difference is worth stating precisely because it is easy to collapse. Tracing asks whether some +human in the design-and-use chain understood the system and was answerable for it. Delivery asks +whether the governing decision **reached the act** [DDD-delivery-02, projected]. An arrangement can +satisfy tracing completely — a named, understanding, answerable designer — and still fail at the act, +because the decision that designer authored was never retrieved when it was needed. That is escape +that presents as governance, and it is a failure a tracing audit passes. + +### 11.5 What is left to claim + +The comparison, in one table. Every row's *what the framework adds* is `projected` in the graph +unless marked otherwise, which means proposed with a declared falsifier and not yet met. + +| Neighbour | The unit | What it already has | What the framework adds | Node | +|---|---|---|---|---| +| Hutchins | people + artefacts in a practice | the arrangement as the cognitive unit | the governing questions — acceptance, assurance, principal | `term:arrangement` | +| Hollnagel & Woods | the joint cognitive system | the arrangement as the unit, **prior to this framework** | source separated from assurance; standing separated from occasioned | `DDD-frame-03`, `DDD-cost-09` | +| Leveson | the control structure | control, authority, constraints, four complete types of unsafe control action | filing separated from delivery, as a standing category | `DDD-delivery-01`, `DDD-delivery-02` | +| Horvitz | the interaction | allocation of initiative, **with a decision criterion the framework lacks** | what the arrangement committed in advance, by level | `term:commitment-level` | +| Bovens | the actor–forum relation | four of five accountability elements; the problem of many hands | authority linkage; the relation as a design-time property | `DDD-frame-08` | +| Matthias | the ascription | the gap, grounded in a control condition | a completeness the gap does not require — **and no claim that this closes it** | `DDD-frame-08` | +| Meaningful human control | the system and its design chain | tracking and tracing, separated | delivery: reaching the act, which tracing does not test | `DDD-delivery-02` | + +**(analysis)** What survives the survey is narrower than the merged version of this paper claimed, +and it is what the review itself proposed: **the contribution is a specific, auditable synthesis of +resolution, assurance, delivery and accountability — not the discovery that sociotechnical +arrangements matter.** Each coordinate has a prior literature, and several of those literatures are +older and better evidenced than this framework. What the framework offers is that the four are +carried in one vocabulary, over one unit, with each proposition attached to a node that states its +status and its falsifier — so a reader can find out what is argued and what is merely proposed +without reading the authors' minds. Whether that synthesis earns its keep is an empirical question, +and §9 is where it would be answered. + +**Two things the survey did not find**, recorded because their absence is part of the result. No +work surveyed carries the filing-versus-delivery distinction as a standing category, and none +separates a resolution's source from its assurance mechanism as orthogonal dimensions. Those are the +two places the framework's difference held up under reading. They are also `projected` claims with +empty evidence fields, and a distinction being unoccupied is not a distinction being correct. + +### 11.6 A force the framework names and has not filed **(analysis)** One piece of positioning belongs here because it names the force the whole framework is about, and it is deliberately unfiled. @@ -1260,14 +1411,23 @@ than against the one that happened to be current. ## References -Locators are marked **verified** where the source was checked directly, and **unverified** where it -could not be. One entry below is unverified, and the reason is stated in the entry rather than left -for a reader to discover. +Locators are marked **verified** where the locator was checked against an authoritative registry — +Crossref for a DOI, Open Library or the publisher's own record for an ISBN, NCBI for a PubMed +record — and **unverified** where it could not be. One entry below is unverified, and the reason is +stated in the entry rather than left for a reader to discover. + +**Verifying a locator is not reading a text, and the two are marked separately.** Where an entry's +substance rests on a secondary characterisation rather than on the primary work, the entry says so. +The distinction was earned rather than assumed: the review this revision answers cited the wrong +volume of a two-volume pair, and a database would have propagated the error. Ashby, W. R. (1956). *An Introduction to Cybernetics*. Chapman & Hall, London. Variety and the bit, §7/7; the Law of Requisite Variety in logarithmic form, §11/7, with the general statement at §11/9. *(verified)* +Bovens, M. (2007). Analysing and Assessing Accountability: A Conceptual Framework. *European Law +Journal* 13(4):447–468, doi:10.1111/j.1468-0386.2007.00378.x. *(verified; primary text read.)* + Brooks, F. P. (1987). No Silver Bullet: Essence and Accidents of Software Engineering. *IEEE Computer* 20(4):10–19, April 1987, doi:10.1109/MC.1987.1663532; reprinted from *Proceedings of the IFIP 10th World Computer Congress*, North-Holland, 1986. Both locators are given because both @@ -1284,9 +1444,45 @@ Goodhart, C. A. E. (1975). Problems of Monetary Management: The U.K. Experience. Monetary Economics*, Volume I. Reserve Bank of Australia, Sydney. Reprinted in *Monetary Theory and Practice: The U.K. Experience*, Macmillan, 1984. *(verified)* +Hollnagel, E. and Woods, D. D. (1983). Cognitive Systems Engineering: New wine in new bottles. +*International Journal of Man-Machine Studies* 18(6):583–600; reprinted *International Journal of +Human-Computer Studies* 51(2):339–356, 1999, doi:10.1006/ijhc.1982.0313. Both locators are given +because both circulate. *(verified; the authors' own abstract read, not the full text.)* + +Hollnagel, E. and Woods, D. D. (2005). *Joint Cognitive Systems: Foundations of Cognitive Systems +Engineering*. CRC Press, Boca Raton. ISBN 0-8493-2821-7, doi:10.1201/9781420038194. **Not to be +confused with the companion volume below**, which reverses the author order and shares the main +title. *(Locator verified against Crossref and Open Library; the volume itself was not obtained, +and this paper's characterisation of it rests on the 1983 paper above and on secondary sources.)* + +Horvitz, E. (1999). Principles of Mixed-Initiative User Interfaces. *Proceedings of CHI '99*, +pp. 159–166. ACM Press, doi:10.1145/302979.303030. *(verified; primary text read.)* + +Hutchins, E. (1995). *Cognition in the Wild*. MIT Press, Cambridge, Massachusetts. ISBN +978-0-262-08231-0; paperback, Bradford Books, 1996, ISBN 978-0-262-58146-2. *(Locator verified +against Open Library; characterised here from the author's own summary of the book and from +secondary sources, not from the full text.)* + Johnson, W. E. (1921). *Logic*, Part I, chapter XI, "The Determinable". Cambridge University Press. *(verified)* +Leveson, N. G. (2011). *Engineering a Safer World: Systems Thinking Applied to Safety*. MIT Press, +Cambridge, Massachusetts. ISBN 978-0-262-01662-9; paperback 2017, ISBN 978-0-262-53369-0; an +open-access edition is published by MIT Press. *(Locator verified against Open Library. STPA's +control structure, process models and four types of unsafe control action are cited here from the +STPA Handbook below, which states them and which was read; the book itself was not obtained.)* + +Leveson, N. G. and Thomas, J. P. (2018). *STPA Handbook*. March 2018, published by the authors and +distributed through MIT's Partnership for Systems Approaches to Safety and Security. Chapter 2, +"How to do Basic STPA": the hierarchical control structure, the generic control loop and process +models under "Modeling the control structure", pp. 22–24; the four types of unsafe control action +and the argument that they are complete, pp. 38 and 38–39. The handbook's chapters carry names +rather than numbered subsections, so page numbers are given. *(verified; primary text read.)* + +Matthias, A. (2004). The responsibility gap: Ascribing responsibility for the actions of learning +automata. *Ethics and Information Technology* 6(3):175–183, doi:10.1007/s10676-004-3422-1. +*(verified; primary text read.)* + Meyer, B. (1992). Applying "Design by Contract". *IEEE Computer* 25(10):40–51. *(verified)* Polanyi, M. (1966). *The Tacit Dimension*. Doubleday & Company, Garden City, New York. Based on the @@ -1295,6 +1491,10 @@ Terry Lectures delivered at Yale University in 1962. *(verified)* Prior, A. N. (1949). Determinables, Determinates and Determinants. *Mind* 58(229):1–20 and 58(230):178–194, in two parts. *(verified)* +Santoni de Sio, F. and van den Hoven, J. (2018). Meaningful Human Control over Autonomous Systems: +A Philosophical Account. *Frontiers in Robotics and AI* 5:15, doi:10.3389/frobt.2018.00015. +*(verified; primary text read.)* + Shannon, C. E. (1948). A Mathematical Theory of Communication. *Bell System Technical Journal* 27(3):379–423 and 27(4):623–656. *(verified)* @@ -1307,6 +1507,12 @@ cites Tesler for the allocation question only, and takes no result from him. Wilson, J. (2023). Determinables and Determinates. *Stanford Encyclopedia of Philosophy*, first published 7 February 2017, substantive revision 18 January 2023. *(verified)* +Woods, D. D. and Hollnagel, E. (2006). *Joint Cognitive Systems: Patterns in Cognitive Systems +Engineering*. CRC Press, Boca Raton. ISBN 0-8493-3933-2, doi:10.1201/9781420005684. **This is the +volume the external review's own locator resolves to**, while the review's prose concerns the 2005 +*Foundations* volume above. *(Locator verified against Crossref and Open Library; the volume itself +was not obtained.)* + --- ## Appendix A. Cited claims, decisions and terms From 616c6ea3d70b0ee079be222e2ebba23d9ed4c515 Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 31 Aug 2026 13:03:51 +0000 Subject: [PATCH 4/9] GATE 3 (draft-pending-ruling): the narrowed claim, the retitle, the rhetoric pass, and the review's small items MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit R-2. Retitled to candidate 1 on Emil's Gate 1 ruling: "Actor-Indexed Determination", with the existing subtitle promoted unchanged. The narrowed claim is stated once, in §1.2, in the review's own terms — a specific, auditable synthesis of resolution, assurance, delivery and accountability over the arrangement as the unit — together with the withdrawal it rests on: the arrangement is not this framework's, the absence the old title asserted does not hold, and §11 credits whose it is. The abstract and §12 carry the same narrowing. RHETORIC. "Two results give the framework its shape" becomes two proposals with declared falsifiers. "Its strongest result" (§6, §12) and "the strongest defensible result" (§5.4) become load-bearing proposals, each labelled projected at the point of use. §6.2's "narrower result" is labelled reported and the difference from observed is stated. The conclusion's established list now says every claim in it is formal, and that established means internally argued and unchallenged, not externally validated. The closing paragraph no longer asserts the framework's value; it states its aim and names the open question. R-6.2 Accountability counts reconciled in §7 with a mapping table. The five-element claim refines term:accountability's third element into stake and sanction path and ADDS authority linkage. A projected claim adding an element to a settled term is a supersession question; the paper flags that canon disagrees with itself about the relation's arity and does not take the ruling. R-6.3 §8.1's provenance assignments become hypotheses with what would overturn each, and the schema row — the one the review challenged — is shown to sit under two provenances at once, which the five-way partition does not adjudicate. §8.5's two escapes become CANDIDATE escapes: an escape is a property of an arrangement, so a walk that did not look everywhere has not established one, and code search, contract tests, telemetry or asking the consuming teams all reach the question the walk assumed missing. R-6.1 §9.5 gains Study 0 — coding reliability first, per dimension and not pooled — ahead of any comparative work. Two admissions follow: the escaped-decision count is circular until the instrument holds, and failure of the umbrella prediction would not falsify the ontology, because the framework's descriptive and predictive claims are separately falsifiable. NOT IN THE PROMPT'S R-6 LIST, delivered and flagged for ruling: the closure ladder's axis mixing, which the triage records as Emil's own Gate 1 ruling at the drafting session. The ladder now ends at constructively closed and is three rungs, all operational; decidability is stated as the logical axis's answer with both directions of non-implication given. Revert if this exceeds the gate. Three stale citations of the old title updated: the reviewer brief, the measure note's paper context, and meta/way-of-working.md. README.md:32 and CHANGELOG.md:303 use the phrase rather than the title and are left alone. All instruments green at the pin: 29 verbatim / 0 failing, 72 nodes / 0 discrepancies, 85 inline status assertions / 0 wrong; the three repository validators pass with the six recorded warnings. Basis: DDD-frame-01; DDD-frame-04; DDD-frame-05; DDD-frame-07; DDD-frame-08; DDD-frame-11; DDD-floor-02; DDD-hyp-04; DDD-dec-26; term:closure; term:accountability; term:attribution; term:answerability; term:liability; term:commitment-level --- meta/way-of-working.md | 2 +- papers/measure-note/measure-paper-context.md | 3 +- papers/paper-a/paper-a.md | 257 +++++++++++++++---- papers/paper-a/reviewer-brief.md | 2 +- 4 files changed, 207 insertions(+), 57 deletions(-) diff --git a/meta/way-of-working.md b/meta/way-of-working.md index 467af94..155f131 100644 --- a/meta/way-of-working.md +++ b/meta/way-of-working.md @@ -180,7 +180,7 @@ Claim node, stated with its falsifier per convention: > or tool maintenance exceeds the audit cost it replaces. **What breaks:** nothing in `core/` — the > tool is a projection; if it fails, the discipline reverts to convention and grep. -**Paper A** (*The Missing Parameter: Actor-Indexed Determination*) is not a paper on this roadmap. It +**Paper A** (*Actor-Indexed Determination*) is not a paper on this roadmap. It is the program's conceptual material. The foundation-revision document is quarried, not completed: its claim-status table and hypothesis structure move into `meta/`; its manuscript prose waits. A synthesis paper may exist eventually — written last, organising results rather than proposing diff --git a/papers/measure-note/measure-paper-context.md b/papers/measure-note/measure-paper-context.md index aa4cba1..28fb64d 100644 --- a/papers/measure-note/measure-paper-context.md +++ b/papers/measure-note/measure-paper-context.md @@ -37,7 +37,8 @@ is as long as its booked content, reported honestly, with ceiling questions take padding, and no trimming of reviewed content to pay for booked additions. When a figure lands past what was last accepted, raise it as a question rather than let it drift. -**Relationship to Paper A.** Paper A (*The Missing Parameter: Actor-Indexed Determination*) states +**Relationship to Paper A.** Paper A (*Actor-Indexed Determination*; retitled at its 2026-08-30 +revision from *The Missing Parameter*, which asserted an absence its survey did not support) states the actor-general principle and cites this result rather than deriving it. This note cites A for the framework and does not restate it. diff --git a/papers/paper-a/paper-a.md b/papers/paper-a/paper-a.md index 1d07de7..69ce3e7 100644 --- a/papers/paper-a/paper-a.md +++ b/papers/paper-a/paper-a.md @@ -1,4 +1,4 @@ -# The Missing Parameter: Actor-Indexed Determination +# Actor-Indexed Determination ### Where consequential choices are resolved, what assures them, and who answers @@ -29,16 +29,23 @@ from the *mechanism* that assures it. This yields a precise account of an **esca consequential resolution that no adequate source-and-assurance combination governs, and whose unacceptable outcomes are not reliably detected. -Two results give the framework its shape. Every completed act's determination demand is discharged -— by a filed decision, an actor's judgment, an arrangement default, or an uncontrolled draw — so -**demand is never unmet, only ungoverned**; and discharge is act-indexed, so governance never -chooses whether demand is supplied, only by what. On the region where the acceptance predicate -closes, that demand is measurable, and its conservation is the chain rule of entropy. Off that -region the measure does not exist and the governance question still does. - -The contribution is not a new species taxonomy of programs, models and humans. It is a framework -for analysing where consequential choices are resolved, how commitments attach, what can be -verified, who bears the residual consequence, and which choices have been left to accident. +Two propositions give the framework its shape, and both are **proposals with declared falsifiers +rather than findings**. Every completed act's determination demand is discharged — by a filed +decision, an actor's judgment, an arrangement default, or an uncontrolled draw — so **demand is +never unmet, only ungoverned**; and discharge is act-indexed, so governance never chooses whether +demand is supplied, only by what. On the region where the acceptance predicate closes, demand is +identified with the entropy of the verdict, and on that identification conservation is the chain +rule of entropy. The identification is a modelling claim and it is unproven; the chain rule is +Shannon's and is certain and empty on its own. Off the closing region the measure does not exist +and the governance question still does. + +**What is claimed is a synthesis, not a discovery.** That consequential work should be analysed at +the level of a sociotechnical arrangement rather than an isolated actor is not this framework's +proposition: it belongs to distributed cognition, to joint cognitive systems, and to systems-theoretic +safety, and §11 credits them. What is offered here is a specific, auditable synthesis of resolution, +assurance, delivery and accountability over that unit — with every load-bearing proposition attached +to a graph node that carries its status and its falsifier, so a reader can tell what is argued from +what is merely proposed. No empirical result is claimed, and none is reported. --- @@ -111,6 +118,20 @@ and computational work. It does not claim that determination demand is a measure general — only on the region where the acceptance predicate closes, where the measure exists and the conservation statement is arithmetic (§4.4). +**And it does not claim the arrangement.** An earlier version of this paper was titled *The Missing +Parameter*, which asserts absence from prior work. The absence does not hold. Analysing +consequential work at the level of an arrangement rather than an isolated actor is central to +distributed cognition, to joint cognitive systems and to systems-theoretic safety, all of which +predate this framework by decades and several of which are better evidenced than it is (§11). The +title was withdrawn on that finding. + +What is claimed instead is narrower and is stated once, here, in the form the paper defends +everywhere else: **a specific, auditable synthesis of resolution, assurance, delivery and +accountability over the arrangement as the unit** — four coordinates each of which has a prior +literature, carried in one vocabulary, with each load-bearing proposition attached to a node +carrying its status and its falsifier. Whether that synthesis is worth having is an empirical +question and §9 is where it would be answered. **(analysis)** + One question sits adjacent to this paper and is deliberately not taken up: software mechanised the half of engineering that names what varies, and never mechanised the half that resolves it — an asymmetry with its own literature and its own burden of survey, which belongs to a different paper. @@ -581,25 +602,29 @@ logically and operationally available and still economically useless. and stakeholders the arrangement is meant to serve. A mechanically executable predicate may be normatively incomplete, contested or gameable. -**A second axis follows, and it is orthogonal to this one.** These four name *which* closure -question is being asked. The next section asks *how strong* the answer is once the question is -fixed as operational. A predicate has a position on both axes at once, and neither reading -constrains the other — so nothing in §5.2 competes with the four kinds, and a reader meeting both -should not look for a contradiction between them. +**A second axis follows, and the earlier version of this paper mixed it with this one.** These four +name *which* closure question is being asked. The next section asks *how strong* the answer is once +the question is fixed as **operational**. A predicate has a position on both axes at once. + +**(analysis)** The external review found the earlier §5.2 ending its operational ladder at *formally +decidable*, which is a property on the **logical** axis and not a stronger operational rung. The +finding is correct and the ladder is repaired below: it ends at constructively closed, and +decidability is stated as what it is — the logical axis's answer, related to the operational one and +not above it. The defect was in this paper's framing and not in canon, which reserves *decidable* +for the formal special case rather than making it the requirement [term:closure, settled]. ### 5.2 How strong is the closure? -**(analysis)** Section 5.1 sorted the questions. This section orders the answers to one of them. +**(analysis)** Section 5.1 sorted the questions. This section orders the answers to **one** of them. Fix the question as operational closure — the framework's own [term:closure] — and predicates still differ in *degree*, along a gradient the framework's material already implies and has not yet -assembled in one place. Four rungs, weakest first. +assembled in one place. **Three rungs, weakest first**, and all three are operational. | Rung | What holds | Where the framework says so | |---|---|---| | **Open** | No acceptance procedure over accessible ground at the declared assurance level. | [term:closure], by negation | | **Verification-closed** | Adequacy can be *evaluated* within declared resource, latency and confidence bounds. | [term:closure] | | **Constructively closed** | The verdict is *computed by rule* from ground available at the act. | measure note §8 | -| **Formally decidable** | An acceptance procedure exists and terminates over the declared domain. | [term:closure]'s reservation | **Open.** No procedure decides acceptability, so there is no verdict function to have entropy about and the measure does not exist [DDD-measure-06, **established**]. Governance is untouched by this: @@ -620,15 +645,25 @@ asks whether adequacy can be *evaluated*; constructive closure asks whether the its entropies are exact and exhaustive rather than sampled precisely because the verdict there is computed rather than searched for (measure note §8). -**Formally decidable.** Placed last and deliberately not at the top. Canon reserves *decidable* for -the formal special case rather than making it the requirement [term:closure], and the reservation is -load-bearing in both directions: any bounded finite domain is decidable by lookup, and a decidable -checker may demand resources no arrangement has. Decidability is therefore a special case of the -ladder rather than its summit, and the rungs that govern deployed arrangements are the middle two. +**And decidability is not a fourth rung.** An earlier version of this paper put *formally decidable* +at the ladder's end, which was an error: decidability is an answer on §5.1's **logical** axis — a +procedure exists and terminates over the declared domain — and it is not a stronger operational +position than constructive closure. The two axes relate, and the relation is worth stating rather +than implying. + +**(analysis)** Neither direction of implication holds. A decidable predicate need not be +operationally closed for any arrangement: any bounded finite domain is decidable by lookup, and a +decidable checker may demand time, memory, permissions or ground no arrangement has. And an +operationally closed predicate need not be decidable: closure asks whether adequacy can be evaluated +within declared bounds *for this arrangement*, and a procedure can meet those bounds on the inputs +the task actually presents without terminating over the declared domain. Canon reserves *decidable* +for the formal special case rather than making it the requirement [term:closure, settled], and the +reservation is what keeps the two axes apart. **The rungs that govern deployed arrangements are the +last two, and they are operational.** #### The retirement this ladder walks past -The third rung approaches a claim the framework killed, and the approach has to be stated carefully +The top rung approaches a claim the framework killed, and the approach has to be stated carefully because the resemblance is real. > **RETIRED — "closed predicates make intelligence unnecessary."** Does not follow from @@ -680,8 +715,10 @@ capability, access to relevant ground, and adversarial pressure. **(analysis)** ### 5.4 Producer independence, and nothing more -The strongest defensible result about closure is narrow, and the framework states its limits inside -the claim rather than in a footnote: +The most defensible proposition the framework offers about closure is narrow, and the framework +states its limits inside the claim rather than in a footnote. It is `projected` — proposed with a +declared falsifier and not met — so what follows is what the framework is prepared to defend, not +what it has shown: > **Under a sound terminating operational checker with complete declared ground, producer identity > is not epistemically necessary for the checked property — and nothing more: not cheap generation, @@ -723,7 +760,8 @@ framework's hypotheses are written to respect that difference (§9) [DDD-hyp-02, ## 6. Actor-indexed irreducibility Brooks's essential/accidental distinction becomes more precise when it is indexed. The framework's -statement of that precision is its strongest result, and it is relational: +statement of that precision is relational, and it is `projected` — the most load-bearing proposal +in the paper, and a proposal: > **The judgment floor is relational: irreducibility is a property of the indexed relation ⟨task, > ground, acceptance relation, tolerance, arrangement, assurance⟩, not of the task alone — the @@ -764,7 +802,9 @@ it does (measure note §8.2). ### 6.2 The capacity mechanism, and its scope -A second and narrower result describes one *generator* of floor rather than the floor itself. +A second and narrower claim describes one *generator* of floor rather than the floor itself. It is +`reported` — exercised by a reproducing computation, which is not the same as observed in the +world: > **H(V|X) bundles judged and escaped demand; cleaving them requires an actor-capacity model, and > residual demand an actor has taken up escapes where it exceeds effective capacity `min(C_hold, @@ -809,6 +849,36 @@ to explain [term:answerability, settled]. **Liability** is exposure to consequen settled]. **Accountability** is the relation that requires all of them together with a persistent principal, authority linkage and a sanction path [term:accountability, settled]. +#### The five elements and the three, reconciled + +The graph carries **two counts** of accountability's elements, and the external review was right +that the paper used both without supplying the mapping. The claim above names five — attribution, +persistent principal, authority linkage, stake, sanction path. The canonical term names three: *"the +attribution of the determination, a persistent answerable party, and a borne consequence"* +[term:accountability, settled; reproduced in full in Appendix A]. The mapping is not one-to-one and +the difference is not cosmetic. + +| `term:accountability` (three, **settled**) | `DDD-frame-08` (five, **projected**) | Relation | +|---|---|---| +| attribution of the determination | attribution | the same element | +| a persistent answerable party | persistent principal | the same element | +| a borne consequence | **stake** and **sanction path** | the five-element version *splits* one into exposure and the mechanism that imposes it | +| — | **authority linkage** | the five-element version *adds* an element the term does not carry | + +So the five-element version **refines** the three-element one on its third element and **adds** a +fourth kind of thing. It does not replace it. + +**(analysis)** Two consequences the paper carries rather than resolves. The split is defensible: +exposure to consequence and a body able to impose one are separable, and an arrangement can have the +first without the second — an executor who will be blamed and no forum that can act. The addition is +the live question. Authority linkage — that the acting party was *entitled* to act — is not in the +settled term, and a `projected` claim that adds an element to a `settled` term is a supersession +question rather than an elaboration. **This paper does not take it.** It flags that the two nodes +disagree about the relation's arity, that the disagreement is in canon and not in this projection, +and that a canon session owes the ruling. Reading the five-element version as settled because the +three-element version is settled would be exactly the status confusion the note on claim status +warns against. + ### 7.1 Executor and principal The executor produces the resolution. The principal answers for it. These are routinely the same @@ -860,15 +930,32 @@ and a human reviewer is available. ### 8.1 Ground and its provenance -The relevant ground is heterogeneous, and §2.4's provenance vocabulary sorts it. The repository's -conventions are **controlled**. Each country's address-format rules are **institutional**, supplied -by postal authorities rather than derived. The current schema is **observed**, and observed *from a -store*, so it is current only as of its last write. Whether a customer record is a business or a -residence is **inferred**. And whether downstream consumers depend on the existing un-normalised -form is, in most real repositories, **missing**. +The relevant ground is heterogeneous, and §2.4's provenance vocabulary sorts it. **Every assignment +below is a hypothesis about this arrangement, to be investigated in it — not a fact the scenario +supplies.** The external review found the earlier version of this section applying its own taxonomy +inconsistently, and the repair is not a better set of assignments but an honest account of what an +assignment is: a claim about where a fact comes from *in a particular arrangement*, which an +investigation can overturn. + +| Ground | Assignment | What would overturn it | +|---|---|---| +| the repository's conventions | **controlled** | the conventions are set outside the team and merely mirrored here, in which case they are institutional | +| each country's address-format rules | **institutional**, supplied by postal authorities | the arrangement maintains its own rule table and enforces it, in which case they are controlled | +| the current schema | **observed**, and from a store | the arrangement owns and enforces the schema, in which case it is controlled — this is the assignment the review challenged, and it turns on who may change it | +| business or residence | **inferred** | the field is declared at signup, or read from an external register, in which case it is neither | +| whether consumers depend on the un-normalised form | **missing** | code search, contract tests, telemetry or asking the consuming teams reaches it — see §8.5 | + +**(analysis)** The schema row is the instructive one. An arrangement that both maintains a schema and +reads it back from a store holds it under two provenances at once, and the taxonomy does not +adjudicate: which one governs depends on whether the acceptance question is *what the schema says* +or *what the schema currently is*. That is a limit of the five-way partition, and the partition is +this paper's analysis rather than canon, ruled ineligible for minting for a related reason +[DDD-dec-26]. The missing slot is not a defect in the taxonomy. It is the finding: an act whose acceptance depends -on ground the arrangement cannot reach has a floor, whatever the executor's capability (§6.1). +on ground the arrangement cannot reach has a floor, whatever the executor's capability (§6.1). **That +too is arrangement-relative** — ground missing to one arrangement is available to another that can +run the search, and the floor moves with it [DDD-floor-02, projected]. ### 8.2 Commitments, by level @@ -915,10 +1002,20 @@ computed per capability rather than per act [DDD-cost-12]. ### 8.5 What the walk exposes, and what it predicts -Two escapes, both invisible to the arrangement's own instruments: the library default on uncovered -countries, and the un-normalised-form question that nothing in the arrangement asks. Both are -consequential, neither is governed, and the framework predicts that both surface as ungoverned -failure modes and as design-review findings [DDD-frame-04]. +Two **candidate** escapes, both invisible to the arrangement's own instruments as described: the +library default on uncovered countries, and the un-normalised-form question that nothing in the +arrangement asks. Both are consequential and neither is governed by anything the walk found, and the +framework predicts that escapes of this kind surface as ungoverned failure modes and as +design-review findings [DDD-frame-04, projected]. + +**(analysis)** *Candidate*, because the review's objection is correct and the framework's own +vocabulary makes it: an escape is a property of an arrangement, not of an outcome, so a walk that +did not look everywhere has not established one. The un-normalised-form question is reachable by +code search, by contract tests, by telemetry on the existing endpoint, or by asking the consuming +teams — and an arrangement that does any of those has ground the walk assumed missing. What the walk +actually shows is weaker and still worth having: **the arrangement's declared instruments do not ask +either question**, and nothing in its ledger would record that they went unasked. Whether a wider +search reaches them is exactly the investigation §8.1 says each provenance assignment needs. The remedy the framework recommends is not more capability. It is declaring the determinables — the uncovered-country behaviour and the preservation question become decisions once the arrangement @@ -1020,6 +1117,48 @@ identity persistence held constant [DDD-hyp-05, notes]. **(analysis)** The design below is the paper's proposal; the preregistration artefact belongs to the study that owns the hypothesis set, and is referenced here rather than asserted. +#### The first study is not a comparison + +The external review's finding was that the five hypotheses are **not yet study-ready**, and it is +correct. H1 bundles evaluability, feedback speed, feedback density, ground access, checker cost and +retry cost into one prediction. H2 bundles delayed consequences, evaluator disagreement, criterion +drift, tacit knowledge and legitimacy. Difficulty and resources are invoked as controls with no +operational definition. Each hypothesis is therefore **a research programme rather than a study**, +and each needs decomposing into preregistered pieces before any of it is run. + +**And none of it can be run first.** Every dependent variable in the design below is read through +the framework's own coding scheme: an arrangement's commitments are coded by level, its ground by +provenance, its predicates by closure rung, its escapes by structured review. If two competent +coders do not agree on those codes, every downstream comparison measures the coders. **The first +study is therefore an instrument study, not a comparison**, and it has to come first: + +**Study 0 — coding reliability and predictive validity of the framework's coding scheme.** +Independent coders, blind to each other and to the hypotheses, code a corpus of real arrangement-task +pairs on five dimensions: commitment level per determinable [term:commitment-level, draft]; ground +provenance (§2.4, this paper's analysis and not canon); closure rung per predicate (§5.2); discharge +mode per resolution; and accountability elements present [DDD-frame-08, projected]. Reported: +inter-rater agreement **per dimension, separately rather than pooled**, since a scheme can be +reliable on closure and unreliable on discharge, and a pooled figure would hide exactly that. Then, +on the codes that survive: whether they predict anything a task-only description does not. + +Two things follow, and both are uncomfortable for the framework. + +*The escaped-decision count is circular until Study 0 passes.* §9's dependent variables include +"escaped-decision count found by structured design review", where the review uses this framework's +criterion. A count produced by the framework cannot also be evidence for it. The count is usable +only as an *inter-rater* quantity — do independent reviewers find the same escapes — until an +independent criterion exists, and the paper does not have one. + +*The umbrella prediction's failure would not falsify the ontology.* If arrangement-level variables +carry no more predictive weight than executor kind [DDD-frame-07, projected], what fails is the +comparative-advantage prediction. The vocabulary — source separated from assurance, filing from +delivery, executor from principal — could still be the right way to describe arrangements while +predicting nothing about which one wins. **The framework's descriptive and predictive claims are +separately falsifiable, and only the predictive ones are what §9 tests.** The review made this +point and the paper had not. + +#### The comparative design, once the instrument holds + *Unit of analysis.* The arrangement-task pair, never the actor. An arrangement is specified by its executor, its commitments by level, its ground channels, its verification mechanisms, its reviewers and its accountable principal (§2.5). @@ -1055,8 +1194,8 @@ with the correction that killed them. IDs are never reused and retired claims ar > producer-independence under verification; generation may still require whatever capability the > search demands. [DDD-frame-09, **retired**] -What survives is the scoped result of §5.4: producer identity is not necessary *for the checked -property and nothing more* [DDD-frame-05]. The retirement is the difference between those two +What survives is the scoped claim of §5.4, itself `projected`: producer identity is not necessary +*for the checked property and nothing more* [DDD-frame-05, projected]. The retirement is the difference between those two statements, and the difference is the whole of §5.3. > **RETIRED — "a better decomposition destroys demand."** Cheaper parts were purchased by a @@ -1142,7 +1281,7 @@ enumeration is `projected` [DDD-frame-01]. The framework indexes both questions (§6.3) and, on the closing region only, measures the second (§4.4). It does not claim their results as instances of its own. -**Formal verification and proof-carrying systems.** The producer-independence result (§5.4) is the +**Formal verification and proof-carrying systems.** The producer-independence claim (§5.4) is the framework's statement of what a sound check buys, deliberately narrowed to the checked property [DDD-frame-05, projected]. The literature's own scope conditions — soundness, ground completeness, checker legitimacy — are carried inside the claim rather than assumed away. @@ -1343,8 +1482,11 @@ A consequential choice becomes easy, hard, checkable, delegable or judgment-heav what the arrangement can observe, what commitments it already contains, what acceptance relation it must satisfy, what tolerance it declares, and what assurance is required. -That is the missing parameter. It is not a new species name for the determiner. It is the -arrangement through which determination is produced and governed [DDD-frame-01]. +That relativity is what the index names. It is not a new species name for the determiner, and it is +not a parameter this paper discovered: the arrangement through which determination is produced and +governed is the unit distributed cognition, joint cognitive systems and systems-theoretic safety +already work in (§11). What the index adds is that the arrangement's coordinates can be enumerated +and moved one at a time [DDD-frame-01, projected]. Once the arrangement is explicit, several distinctions sharpen. A commitment may attach to an outcome, a policy or a principal, and the three compose [DDD-frame-02]. A resolution's source is @@ -1363,7 +1505,8 @@ claims rather than either alone: every completed act's demand is discharged by o [DDD-frame-15], and discharge is act-indexed, so governance never chooses whether demand is supplied, only by what [DDD-frame-16]. -And its strongest result is relational: +And its most load-bearing proposal is relational — a proposal, with a falsifier, and without +evidence: > **The judgment floor is relational: irreducibility is a property of the indexed relation ⟨task, > ground, acceptance relation, tolerance, arrangement, assurance⟩, not of the task alone …** @@ -1374,9 +1517,13 @@ from generation cost; that conservation on the closing region is the chain rule; exists exactly where the acceptance predicate closes; that the parts of a decomposition become trivial only when the decomposition already encodes the verdict. Projected, with falsifiers and without evidence: the index itself, the commitment levels, the discharge chain, the relational -floor, the accountability relation, and every one of the five hypotheses. This paper states a -framework whose central claims are proposals, and the honest form of that statement is this -sentence rather than a hedge distributed through the argument. +floor, the accountability relation, and every one of the five hypotheses. **Every claim in the first +list is `formal`** — arithmetic, or a derivation from it — so what is settled here is arithmetic, and +the modelling claims that give the arithmetic its meaning are all in the second list. **And +`established` in this graph means internally argued and unchallenged, not externally validated; +nothing in either list has been checked against the world.** This paper states a framework whose +central claims are proposals, and the honest form of that statement is these sentences rather than a +hedge distributed through the argument. The empirical question is not whether models replace people wherever a predicate closes. It is how comparative advantage shifts as operational evaluability, generation cost, ground access, feedback @@ -1385,11 +1532,13 @@ falsification, and it matches engineering practice, where the practical winner i isolated actor and usually a composition of commitments, generators, checks, reviewers and principals. -Decision-Driven Design is the engineering corpus from which this program was abstracted. Its value -is not that it proves a universal conservation principle or an exact boundary between kinds of -determiner. Its value is that it makes consequential choices auditable: where they are resolved, -what they are resolved against, how they are checked, who answers for them, and which ones have -been left to accident. +Decision-Driven Design is the engineering corpus from which this programme was abstracted. Its aim +is not to prove a universal conservation principle or an exact boundary between kinds of determiner, +and it has proved neither. Its aim is to make consequential choices auditable: where they are +resolved, what they are resolved against, how they are checked, who answers for them, and which ones +have been left to accident. **Whether the vocabulary offered here does that better than the +vocabularies it was assembled from is the open question**, and §11 names the ones it was assembled +from. --- diff --git a/papers/paper-a/reviewer-brief.md b/papers/paper-a/reviewer-brief.md index d21948f..28d9f7d 100644 --- a/papers/paper-a/reviewer-brief.md +++ b/papers/paper-a/reviewer-brief.md @@ -1,4 +1,4 @@ -# Reviewer brief — *The Missing Parameter: Actor-Indexed Determination* +# Reviewer brief — *Actor-Indexed Determination* *Companion to the paper. One page. The first brief in this series was written for an information theorist, and asked whether an identification was sound. This one is written for the framework's From 7c9e2faf3098bf032ba727811cb1110170bb5158 Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 31 Aug 2026 13:40:17 +0000 Subject: [PATCH 5/9] GATE 4 step 1: the pin advance predicted, before graph/upstream.yaml moves MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Committed ahead of the operation so the advance is observed rather than assumed, per the DDD-dec-29 pattern. graph/upstream.yaml still reads ref: v5.9.0 at this commit. Predicted: E12 0 · W5 exactly 1 (DDD-frame-15, projected -> retired) · W6 exactly 6 (DDD-cost-09 region; DDD-delivery-01, DDD-frame-02 statement; DDD-frame-15 statement and region; term:delivery and term:residual-discretion canonical_md) · W7 1, unchanged. 61 of 67 pins still. The three the migration seed predicted for this advance — term:delivery, DDD-cost-09, DDD-delivery-01 — appear exactly. The other three are Phase 1a's repairs, which the seed does not list because they are not ground-migration work. None of the seed's seven W1 W6 fires here, and if one does the prediction is wrong and the advance stops. DDD-measure-06 retires at v5.12.0 and fires NOTHING, because it is not pinned — the node forcing the largest prose repair in this revision is invisible to the pin instrument. Pinning it and its two successors is proposed at the advance as a ruling, not a mechanical step. Basis: DDD-dec-29; DDD-dec-28; DDD-agent-01; DDD-frame-15; DDD-frame-02; DDD-cost-09; DDD-delivery-01; term:delivery; term:residual-discretion --- .../gate4-prediction.md | 72 ++++++++++++++++++ .../predict-pin-advance.py | 73 +++++++++++++++++++ 2 files changed, 145 insertions(+) create mode 100644 meta/sessions/2026-08-30-paper-a-revision/gate4-prediction.md create mode 100644 meta/sessions/2026-08-30-paper-a-revision/predict-pin-advance.py diff --git a/meta/sessions/2026-08-30-paper-a-revision/gate4-prediction.md b/meta/sessions/2026-08-30-paper-a-revision/gate4-prediction.md new file mode 100644 index 0000000..2f1ab92 --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/gate4-prediction.md @@ -0,0 +1,72 @@ +# GATE 4 — the pin advance, predicted before it is run + +**Status: prediction, committed before the operation.** `graph/upstream.yaml` still reads +`ref: v5.9.0` at this commit. Nothing has been advanced. This document exists so the advance can be +**observed rather than assumed**, which is the `DDD-dec-29` pattern: state the firing, advance the +ref, re-instrument, and compare. + +Computed by `predict-pin-advance.py` in this directory, which reads the pin list at the current ref +and re-implements `validate-core-order.py`'s own `pinned_content_digest` — sha256 over +`statement`, `region` and `canonical_md`, joined with `\x00`, with `status` deliberately excluded +because `W5` already instruments it. + +--- + +## The prediction + +**`E12` — 0.** Every one of the 67 pinned ids still exists at `v5.12.0`. Verified independently at +Gate 1 by a different route: `gen-appendix.py` resolved all 72 nodes the paper cites, and would +have exited had one been missing. + +**`W5` basis loss — exactly 1.** + +| id | pinned at | at `v5.12.0` | +|---|---|---| +| `DDD-frame-15` | `projected` | **`retired`** | + +`DDD-measure-06` also retires at `v5.12.0` and **will not fire `W5`, because it is not pinned.** +That is worth stating plainly: the node whose retirement forces the largest prose repair in this +revision is invisible to the pin instrument, and Gate 1 found it only because a sweep was written +for inline status assertions. A claim a projection leans on without pinning is a claim that can +move silently — `DDD-agent-01` applied to this repository's own instruments. **Pinning +`DDD-measure-06`, `DDD-measure-16` and `DDD-measure-17` is proposed at this advance** and is a +ruling, not a mechanical step. + +**`W6` content drift — exactly 6**, on the three fields the digest covers. + +| id | field that moved | from → to | +|---|---|---| +| `DDD-cost-09` | `region` | `fecb317f…` → `f6cc24f0…` | +| `DDD-delivery-01` | `statement` | `67377ef2…` → `1340d483…` | +| `DDD-frame-02` | `statement` | `f45492ee…` → `0002f60b…` | +| `DDD-frame-15` | `statement`, `region` | `6d14509d…` → `2a65878d…` | +| `term:delivery` | `canonical_md` | `1e7b2e4d…` → `c56909e5…` | +| `term:residual-discretion` | `canonical_md` | `77284b14…` → `9bbe31e3…` | + +**`W7` shadowed ids — 1, unchanged.** `term:maturation`, declared against `DDD-dec-21`. It has +stood at every advance since and this one does not touch it. + +**Unchanged pins: 61 of 67.** + +--- + +## Reconciliation against the migration seed's own prediction + +`meta/migration-plan-ground.md` predicts the next advance past both sessions: *"Add three that this +session created and did not fire, because the pin stays at `v5.9.0`: `term:delivery`, +`DDD-cost-09`, `DDD-delivery-01`."* + +**Those three appear, exactly.** The other three — `DDD-frame-02`, `DDD-frame-15` and +`term:residual-discretion` — are the Phase 1a session's repairs and are not ground-migration work, +which is why the seed does not list them. The seed also names **seven** W6 that fire when W1 +executes in canon (`term:actor`, `term:arrangement`, `term:capability`, `term:capacity`, +`term:closure`, `term:judgment`, `term:residual-discretion`). **None of those seven fires here**, +because W1 renames nothing in the principle repository in this session: this session carries W1 in +the two manuscripts only. If any of the seven fires at this advance, the prediction is wrong and +something has moved that this session does not understand. + +## What would make this prediction wrong, and what happens then + +Any count other than `E12` 0 · `W5` 1 · `W6` 6 · `W7` 1, or any id outside the six named above. +If that happens the advance stops and the difference is reported at this gate rather than +reconciled afterwards. **The prediction is the instrument here; the advance is only the reading.** diff --git a/meta/sessions/2026-08-30-paper-a-revision/predict-pin-advance.py b/meta/sessions/2026-08-30-paper-a-revision/predict-pin-advance.py new file mode 100644 index 0000000..660898c --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/predict-pin-advance.py @@ -0,0 +1,73 @@ +"""Predict the pin advance's firing WITHOUT touching graph/upstream.yaml. + +Reads the pin list at the CURRENT pin, and computes, for each pinned id, the status and +content digest at both refs, using validate-core-order.py's own hashing so the prediction is +made with the instrument's arithmetic rather than an approximation of it. +""" +import hashlib, subprocess, sys, yaml + +REPO='/home/user/actor-indexed-determination' +OLD, NEW = 'v5.9.0', 'v5.12.0' + +def show(ref, p): + r = subprocess.run(['git','-C',REPO,'show',f'{ref}:{p}'], capture_output=True, text=True) + return None if r.returncode else r.stdout + +def digest(o): + parts=[o.get('statement',''), o.get('region',''), o.get('canonical_md','')] + return 'sha256:'+hashlib.sha256('\x00'.join(parts).encode()).hexdigest() + +def graph(ref): + objs={} + for d in ('core/claims/','core/decisions/'): + for f in subprocess.run(['git','-C',REPO,'ls-tree','-r','--name-only',ref,d], + capture_output=True,text=True).stdout.split(): + if f.endswith('.yaml'): + y=yaml.safe_load(show(ref,f)) + objs[y['id']]={'status':str(y.get('status','')), + 'canonical_md':(y.get('canonical_md') or '').strip(), + 'statement':(y.get('statement') or '').strip(), + 'region':(y.get('region') or '').strip()} + for t in yaml.safe_load(show(ref,'core/graph/terms.yaml'))['terms']: + objs[t['id']]={'status':str(t.get('status','')), + 'canonical_md':(t.get('canonical_md') or '').strip(), + 'statement':(t.get('statement') or '').strip(), + 'region':(t.get('region') or '').strip()} + return objs + +g_old, g_new = graph(OLD), graph(NEW) +pins = yaml.safe_load(open('graph/upstream.yaml'))['upstream']['pins'] +print(f'{len(pins)} pins in graph/upstream.yaml at ref {OLD}\n') + +e12=[]; w5=[]; w6=[]; ok=0 +for p in pins: + i=p['id'] + if i not in g_new: e12.append(i); continue + if p.get('status_at_pin') != g_new[i]['status']: + w5.append((i, p.get('status_at_pin'), g_new[i]['status'])) + d_new = digest(g_new[i]) + if p.get('content_hash') != d_new: + w6.append((i, p.get('content_hash'), d_new, + 'statement' if g_old.get(i,{}).get('statement')!=g_new[i]['statement'] else '', + 'region' if g_old.get(i,{}).get('region')!=g_new[i]['region'] else '', + 'canonical_md' if g_old.get(i,{}).get('canonical_md')!=g_new[i]['canonical_md'] else '')) + else: ok+=1 + +print(f'E12 (pinned id gone at {NEW}): {len(e12)}', e12 or '') +print(f'\nW5 (status moved since the pin): {len(w5)}') +for i,a,b in w5: print(f' {i}: {a} -> {b}') +print(f'\nW6 (content moved since the pin): {len(w6)}') +for i,a,b,*f in w6: + print(f' {i}: {a[:23]}... -> {b[:23]}... fields: {", ".join(x for x in f if x)}') +print(f'\nunchanged pins: {ok}') + +# W7: local terms shadowing upstream +import pathlib +local=set() +lt = pathlib.Path('core/graph/terms.yaml') +if lt.exists(): + for t in (yaml.safe_load(lt.read_text()) or {}).get('terms',[]) or []: + local.add((t['id'], t.get('shadows_upstream'))) +sh=[(i,s) for i,s in local if i in g_new] +print(f'\nW7 (local term ids shadowing the upstream registry at {NEW}): {len(sh)}') +for i,s in sh: print(f' {i} shadows_upstream: {s or "NOT DECLARED"}') From 2b26f25a7cf58a524583f6036a383c189b2fe146 Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 31 Aug 2026 13:50:52 +0000 Subject: [PATCH 6/9] GATE 4 (draft-pending-ruling): the pin advanced and verified, the five repairs, Appendix A regenerated, the supplement split, and W1 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit THE ADVANCE HELD ITS PREDICTION IN EVERY LIMB. Predicted and committed at 7c9e2fa, before graph/upstream.yaml moved: E12 0, W5 exactly 1, W6 exactly 6, W7 1 unchanged. Observed with ref advanced and hashes deliberately NOT re-instrumented: "67 pins resolved, 1 basis-loss, 6 content-drift, 1 shadowed id" — every id and every hash the one written down beforehand. After re-instrumentation: 71 pins, 0/0/1. The advance is DDD-dec-34. FOUR PINS ADDED, and it is a finding rather than housekeeping. DDD-measure-06 retires at v5.12.0 from established, forces the largest repair in this revision, and FIRED NOTHING — W5 only sees ids that are pinned. DDD-measure-06, DDD-measure-16, DDD-measure-17 and DDD-frame-17 are now pinned. THE FOUR QUOTATIONS, all repaired: DDD-frame-15's statement (retired; §4.1 now carries DDD-frame-17's three values, and says the review killed the four modes), DDD-frame-02's amended clause, term:residual-discretion's amended text with the cryptographic-hash separation, and the compact form — the cheapest, exactly as predicted: it survives verbatim in DDD-frame-17's closing clause, so the citation moved and no prose did. AND THE FIFTH, which no checker saw. DDD-measure-06's biconditional is replaced at six sites by the three separated conditions — existence, availability, estimability — with closure governing only availability. §4.4 records the provenance: the measure note had already conceded both failure directions, canon was never updated, so this projection was stronger than the note sharing its graph, and an external review found it rather than an instrument. APPENDIX A regenerated wholesale against the tag, never hand-edited: 72 nodes and 13 discrepancies before, 75 nodes and 0 after, idempotent across three runs byte-compared. The independent re-read confirms it. THE SUPPLEMENT SPLIT on the ruled boundary — about the graph moves, about the world or about warrant stays. paper-a-supplement.md carries the reproduced nodes, the pin, the filing statuses, the arity finding, the W1 record and Reproduction. Every projected-and-empty sentence stayed in the paper. gen-appendix.py and check-appendix.py learned a fourth argument so citations are read from the manuscript and rendered into the supplement; both carry the new defect history, including the failure mode where passing the supplement as the manuscript renders an empty appendix and exits 0. check-status.py ships as the third checker (GATE 2 ruling), with its own defect history and its origin recorded as its warrant. W1: 15 moved, 16 deferred, of the 31 in the two manuscripts. Every deferral verified untouched by explicit fragment check. §5.4's heading renames with one clause naming the asset it now diverges from — flagged, not ruled. Checkers at v5.12.0: 30 verbatim / 0 failing · 93 status assertions / 0 wrong · 75 nodes / 0 discrepancies. Repository gates green with the six recorded warnings. Basis: DDD-dec-34; DDD-dec-29; DDD-dec-28; DDD-agent-01; DDD-frame-02; DDD-frame-15; DDD-frame-17; DDD-measure-06; DDD-measure-16; DDD-measure-17; DDD-measure-12; term:residual-discretion; term:delivery; term:verdict; term:accountability; DDD-frame-08; DDD-dec-26 --- core/decisions/DDD-dec-34.yaml | 123 +++++ graph/upstream.yaml | 54 +- meta/successor-items-paper-a-revision.md | 139 +++++ papers/measure-note/measure-note.md | 36 +- papers/paper-a/check-appendix.py | 21 +- .../paper-a/check-status.py | 52 +- papers/paper-a/gen-appendix.py | 37 +- papers/paper-a/paper-a-supplement.md | 268 ++++++++++ papers/paper-a/paper-a.md | 476 +++++++++--------- 9 files changed, 905 insertions(+), 301 deletions(-) create mode 100644 core/decisions/DDD-dec-34.yaml create mode 100644 meta/successor-items-paper-a-revision.md rename meta/sessions/2026-08-30-paper-a-revision/status-sweep.py => papers/paper-a/check-status.py (51%) create mode 100644 papers/paper-a/paper-a-supplement.md diff --git a/core/decisions/DDD-dec-34.yaml b/core/decisions/DDD-dec-34.yaml new file mode 100644 index 0000000..67e2d11 --- /dev/null +++ b/core/decisions/DDD-dec-34.yaml @@ -0,0 +1,123 @@ +format: 1 +id: DDD-dec-34 +kind: decision +statement: | + The upstream pin advances v5.9.0 -> v5.12.0 across three releases in one act, six pinned objects are re-instrumented and four unpinned ones are added, and the advance's firing is predicted in a committed file before graph/upstream.yaml is touched — because the node whose retirement forces this revision's largest repair, DDD-measure-06, fired nothing at all: it was never pinned. +principal: Emil +made: 2026-08-30, Paper A revision (GATE 3 ruling to proceed; the advance executed and verified at GATE 4) +resolution: | + The advance, what it fired, and the finding that changes what gets pinned. + + ## Why three releases at once + + The pin was HELD at v5.9.0 by ruling, so canon could be repaired without dragging the projection + behind it. Three releases landed against that hold: v5.10.0 (the Phase 1a session — the discharge + partition and the measure boundary), v5.11.0 (retirement provenance and the validator) and + v5.12.0 (definitions that were never in the registry, and a collision the store won). The hold + was correct and it is discharged here, in one advance, by the paper it was protecting. + + ## Predicted before, verified after + + Stated in this order so the verification can disconfirm it, per DDD-dec-28 and DDD-dec-29, and + COMMITTED AHEAD OF THE OPERATION at + meta/sessions/2026-08-30-paper-a-revision/gate4-prediction.md, with the instrument that computed + it beside it (predict-pin-advance.py, which re-implements validate-core-order.py's own + pinned_content_digest rather than approximating it). + + PREDICTED FIRING: + E12 0 — all 67 pinned ids still exist at v5.12.0 + W5 exactly 1 — DDD-frame-15, projected -> retired + W6 exactly 6 — DDD-cost-09 (region); DDD-delivery-01 (statement); DDD-frame-02 + (statement); DDD-frame-15 (statement and region); term:delivery (canonical_md); + term:residual-discretion (canonical_md) + W7 1, unchanged — term:maturation, declared against DDD-dec-21 + and 61 of 67 pins unmoved. + + OBSERVED, at ref v5.12.0 with the hashes NOT yet re-instrumented: + + 67 pins resolved against the pinned ref, 1 basis-loss, 6 content-drift, 1 shadowed id(s) + + Every id and every hash was the one written down beforehand. **The prediction held in every + limb.** The two-stage method is the point: had the hashes been advanced in the same edit as the + ref, the drift count would have read 0 throughout, the pass and the skip would have produced + identical output, and there would have been nothing to check the prediction against. + + After re-instrumentation: 71 pins resolved, 0 basis-loss, 0 content-drift, 1 shadowed id. + + ## Reconciled against the migration seed's own prediction + + meta/migration-plan-ground.md predicted three W6 for the next advance past both sessions — + term:delivery, DDD-cost-09, DDD-delivery-01. Those three appear exactly. The other three are the + Phase 1a session's repairs and are not ground-migration work, which is why the seed does not + list them. The seed also names seven W6 that fire when W1 executes IN CANON; none fired, because + this session carries W1 in the two manuscripts only and renames nothing upstream. + + ## The finding: four pins added, and why it is a finding + + DDD-measure-06 retires at v5.12.0 from established. It forces the largest prose repair in this + revision — the paper asserted its biconditional as **established** in three places, and the + companion measure note had already conceded both failure directions without canon being updated + to match, so this projection was stronger than the note sharing its graph. + + **It fired nothing.** W5 only sees ids that are pinned, and DDD-measure-06 was not. Both existing + checkers passed all three lines, because none is a block quotation and none is an appendix row. + The gap was found by a sweep written at this session's GATE 1 for inline status assertions, now + shipped as papers/paper-a/check-status.py. + + That is DDD-agent-01 applied to this repository's own apparatus: a basis cited without a pin is a + basis that can move silently. Four nodes are therefore pinned at this advance — DDD-measure-06, + its two successors DDD-measure-16 and DDD-measure-17, and DDD-frame-17, which supersedes + DDD-frame-15. Three of the four could not have been pinned earlier: they do not exist at v5.9.0 + and pinning them there would have fired E12. + + ## The other steps, and their results + + - **The four failing quotations**, all repaired and verified: DDD-frame-15's statement (retired, + the four modes replaced by DDD-frame-17's three values), DDD-frame-02's (amended clause), + term:residual-discretion's (amended, the cryptographic-hash separation), and the compact form + in the conclusion — the cheapest, since the compact form survives verbatim in DDD-frame-17's + closing clause, so the citation moved and no prose did. check-quotations.py at v5.12.0: + **30 verbatim, 0 disclosed-partial, 0 failing.** + - **Appendix A regenerated wholesale** against the tag, never hand-edited, and re-read by the + independent script: **75 nodes rendered, 75 cited, 0 discrepancies**, up from 72 nodes and 13 + discrepancies before regeneration. Idempotent across three runs, byte-compared. + - **check-status.py at v5.12.0: 93 inline status assertions, 0 wrong**, from 4 wrong before the + repair. + - **The declared ref** advanced in the manuscript's front matter, which is now the supplement's + S2. + + ## What this act does not do + + It does not merge anything; Emil ratifies. It does not file, amend or retire any claim in either + repository — the two retirements this revision carries were made upstream at v5.10 and are only + projected here. It does not rule on the accountability arity finding (S4 of the supplement), + which is routed as a successor item rather than ridden in on a pin advance. +basis: + - DDD-dec-29 + - DDD-dec-28 + - DDD-dec-10 + - DDD-agent-01 + - DDD-frame-15 + - DDD-frame-17 + - DDD-measure-06 + - DDD-measure-16 +notes: | + Numbering: decisions share one id space across the principle repository and this one. 33 is the + highest before this file (upstream, at v5.12.0), so this advance is 34. + + Filed at the advance rather than staged [PROPOSED], because the tag it pins already existed: + v5.12.0 resolves to 81f6929, verified as the principle repository's default-branch head before + anything moved. The DDD-dec-28/29 near-miss — an instruction to bump arriving before the tag was + cut — does not apply here, and the file lands ratified rather than being tidied later. + + The ordering constraint from DDD-dec-29 still binds and was followed: the ref written into + graph/upstream.yaml is also written into a published manuscript, so pinning a branch during + staging would put a branch sha into a paper. Tag first, always. + + ## What the two-stage verification cost, and why it is worth it + + Nothing but one extra checker run. The alternative — advance ref and hashes together — produces + a green run whether or not the prediction was right, which is presumed discharge in its exact + form: the artefact recording the pass identical to the artefact recording the skip. Three + advances have now been verified this way (DDD-dec-28, DDD-dec-29, this one) and the prediction + has held each time, which is an argument for keeping the method rather than for dropping it. diff --git a/graph/upstream.yaml b/graph/upstream.yaml index dbbfbb9..2c33826 100644 --- a/graph/upstream.yaml +++ b/graph/upstream.yaml @@ -69,7 +69,24 @@ upstream: # basis-loss, EXACTLY ONE content-drift — term:floor, daf43e07... -> 917f7e4d..., both hashes # as predicted — 0 status movement, the one standing W7 unchanged. The prediction held. No # pins added: v5.9.0 mints nothing. - ref: v5.9.0 # tag, never a branch + # Advanced v5.9.0 -> v5.12.0 at the Paper A revision (2026-08-30), on Emil's GATE 3 ruling; + # the advance is DDD-dec-34 here. Three releases at once, because the pin was HELD at v5.9.0 by + # ruling so canon could move without dragging the projection — v5.10.0 (Phase 1a), v5.11.0 + # (retirement provenance) and v5.12.0 (definitions and the store collision) all landed against + # a projection deliberately left behind. Predicted firing, stated before it ran and committed + # ahead of the operation at meta/sessions/2026-08-30-paper-a-revision/gate4-prediction.md: + # E12 0 - W5 exactly 1 (DDD-frame-15, projected -> retired) - W6 exactly 6 (DDD-cost-09 region; + # DDD-delivery-01 and DDD-frame-02 statement; DDD-frame-15 statement and region; term:delivery + # and term:residual-discretion canonical_md) - W7 1, unchanged. Verified at the bump, by + # advancing ref first and re-instrumenting the hashes second so the firing could be observed + # rather than assumed: 67 pins resolved, 1 basis-loss, 6 content-drift, 1 shadowed id, every id + # and every hash the one written down beforehand. The prediction held in every limb. After + # re-instrumentation: 71 pins, 0 basis-loss, 0 content-drift, 1 shadowed id. + # FOUR PINS ADDED, and the reason is a finding rather than housekeeping: DDD-measure-06 retires + # at v5.12.0 and fired NOTHING, because W5 only sees ids that are pinned. Paper A asserted it as + # **established** in three places and both of its checkers passed those lines. The gap was found + # by a sweep written for inline status assertions, now shipped as papers/paper-a/check-status.py. + ref: v5.12.0 # tag, never a branch pins: # basis claims for DDD-dec-04 (the split decision itself rests on these) - id: DDD-frame-01 @@ -123,7 +140,7 @@ upstream: # the assurance-locus conversion core/13 §5's amortisation corollary rests on (GATE B carve) - id: DDD-cost-09 status_at_pin: projected - content_hash: sha256:fecb317fb0b76a969b7f8f5eebdbfc94ea628faf75deb83142de7efb0dd4c013 + content_hash: sha256:f6cc24f0f406dea4f43785106310669ed81a016202165c0d1c739df31b091ffe # the routing layer core/15 exercises and core/14's register consumes (Wave 2, 2026-08-10) - id: term:capability status_at_pin: settled @@ -163,10 +180,10 @@ upstream: # 2026-08-15; staged with the v5.5.0 advance, DDD-dec-18) - id: term:delivery status_at_pin: draft - content_hash: sha256:1e7b2e4d5df5dad2869d966abe94f5d5677a4a2f4d3e0ff35356534f382ecac2 + content_hash: sha256:c56909e53c93d1cfbf857df605df91607e75176ca146f07731386ef9d166e46e - id: DDD-delivery-01 status_at_pin: projected - content_hash: sha256:67377ef2294de814afd95026f2e7ad0bdb768516696395d05f1ed1645aa6d804 + content_hash: sha256:1340d483e45ad5db10acb754aa335453f719f01e8126c3ee0f028f8c5545a73d # Basis added by the freight session (2026-08-17): four upstream objects gained basedOn # edges from decisions filed here — DDD-dec-20 (the session convention), DDD-dec-22 (the # pin/shadow instrumentation) and DDD-dec-23 (ruling 16). Found by the GATE 6 basis-impact @@ -242,7 +259,7 @@ upstream: content_hash: sha256:5c146793529821d24b0b45137f1d8e7893acd8c6392b3694d7e145078c9d5e3e - id: DDD-frame-02 status_at_pin: projected - content_hash: sha256:f45492eedfaf5dc3e0a92776fb71b61639ac1941486816200b01894ba807f04a + content_hash: sha256:0002f60b446fc9d6a5561f21e61fb10472b3b209a416b4ed10216e37721fdbb3 - id: DDD-frame-13 status_at_pin: projected content_hash: sha256:4add58e0f47c0ff22bb752d6b78aef7f591a14b17d610cdb06677439e0b33276 @@ -274,8 +291,8 @@ upstream: status_at_pin: settled content_hash: sha256:b3d940d3167c8e9e18667a224a13d98bad17509844930b3be9e49834014aa19e - id: DDD-frame-15 - status_at_pin: projected - content_hash: sha256:6d14509db3daaaf1773b6e6d24b6d6b742da2c7fa4f5f6bf18f2286a970df3f9 + status_at_pin: retired + content_hash: sha256:2a65878dfdd95895ef91819fe4545ed125e542346eae2ed9b5e91e50e517d58b - id: DDD-frame-04 status_at_pin: projected content_hash: sha256:eec0a4b49c4f996f68a8b83281e0990b829efd318964f803a16452fb4d32a0d3 @@ -304,4 +321,25 @@ upstream: content_hash: sha256:7714da9eb209a6839b862a36c0aecd48d411f6af18cab4b1acaa7a1dea032167 - id: term:residual-discretion status_at_pin: draft - content_hash: sha256:77284b1467c8a100654279d5a3c86970c791badeb1d891e60f9618e13dc1541e + content_hash: sha256:9bbe31e3a02fd3cee70ad42d560268943cd136aa1c0d389ec081d7af0c579340 + # [PROPOSED] Paper A revision (2026-08-30), added at the v5.12.0 advance. Four nodes the + # manuscript leans on that were NOT pinned, found by the Gate 1 status sweep rather than by + # the instrument: DDD-measure-06 retires at v5.12.0 and fired NOTHING, because W5 only sees + # ids that are pinned. The paper asserted it as **established** in three places and both + # existing checkers passed those lines. That is DDD-agent-01 about this repository's own + # apparatus — a basis cited without a pin is a basis that can move silently — and the four + # are pinned here so the next movement is instrumented rather than swept for. They could + # not be pinned before this advance: DDD-frame-17, DDD-measure-16 and DDD-measure-17 do not + # exist at v5.9.0 and pinning them there would have fired E12. + - id: DDD-frame-17 + status_at_pin: projected + content_hash: sha256:bf8d2c699946fd388ee10af09a0e2441fde92d5d69c8e5e847a2e369ca23248d + - id: DDD-measure-06 + status_at_pin: retired + content_hash: sha256:a7ae6e7ad5ccd0afd3bc35c5fd5ce061532b4f99f4d352464f9cc5b78ee01c39 + - id: DDD-measure-16 + status_at_pin: established + content_hash: sha256:2d763d1adf9e1c68c88d1f5c3e30550a5079e274d84e688c31f5f5bb51f1d94e + - id: DDD-measure-17 + status_at_pin: projected + content_hash: sha256:164a18072145bc4a0cdb3c6e1ca8890fc64c1b36396c73be5ca8bd45c66114e5 diff --git a/meta/successor-items-paper-a-revision.md b/meta/successor-items-paper-a-revision.md new file mode 100644 index 0000000..3fcbe57 --- /dev/null +++ b/meta/successor-items-paper-a-revision.md @@ -0,0 +1,139 @@ +# Successor items — Paper A's revision (2026-08-30) + +Items raised at the Paper A revision and deliberately **not** taken there. Each names the node or +document it lands against and the reason it was not done in-session. **Nothing here is filed canon.** + +The session was projection work: the paper may not introduce claims, so every finding it made about +canon has to leave as an item rather than as an edit. That is the whole reason this file exists. + +--- + +## 1. `term:accountability` and `DDD-frame-08` disagree about the relation's arity + +**Ruled at GATE 3 (Emil): the paper flags, and a canon session gets the issue — both, not either.** + +The graph carries two counts of accountability's elements. + +- `term:accountability`, **settled**: *"attribution of the determination, a persistent answerable + party, and a borne consequence"* — three. +- `DDD-frame-08`, **projected**: *"attribution, persistent principal, authority linkage, stake, + sanction path"* — five. + +The mapping, worked in the paper's §7 and reproduced here so the item stands alone: + +| `term:accountability` (three, settled) | `DDD-frame-08` (five, projected) | Relation | +|---|---|---| +| attribution of the determination | attribution | same element | +| a persistent answerable party | persistent principal | same element | +| a borne consequence | **stake** and **sanction path** | the five-element version *splits* one into exposure and the mechanism that imposes it | +| — | **authority linkage** | the five-element version *adds* an element the term does not carry | + +**The split is defensible; the addition is the question.** Exposure to consequence and a body able +to impose one are separable, and an arrangement can have the first without the second. But a +`projected` claim that adds a fourth kind of thing to a `settled` term is a **supersession question +rather than an elaboration**, and it has been carried as an elaboration since both were filed. + +**Why it was not taken here.** A projection may not rule on canon, and this one did not. The paper +reports the disagreement, gives the mapping, and says explicitly that reading the five-element +version as settled because the three-element version is settled would be the status confusion the +review named. + +**What a canon session owes.** Either amend `term:accountability` to carry authority linkage — in +which case a settled term moves and every projection pinning it fires W6 — or scope `DDD-frame-08`'s +fifth element as an addition the term deliberately does not make, with the reason stated. **Not +both, and not neither.** The cheap moment is before the repositories are public, because the term is +`settled` and an outside reader will read the count as canonical. + +**Bearing on the external review.** Its §10.2 asked for exactly this mapping and asked whether the +five refines or replaces the three. The paper now answers *refines-and-adds*. It does not answer +whether the addition is licensed, and that is this item. + +--- + +## 2. W1's remaining surface, and the counts the charter got wrong + +**Reported at GATE 1, corrected in the migration seed on Emil's ruling.** + +The session's charter said 88% of W1 lives in the two papers. Against the committed classification: +**88 is the corpus-wide mutable total, and the two manuscripts hold 31 of it — 35%.** The migration +seed is corrected. **W1's remaining surface after this session is 73 of 88**, and it is not in the +papers. + +Sixteen of the 31 were deferred here, in three classes, and each class is a rule rather than a +judgement call: + +- **generated or hand-carried node tables** (8) — the words are canon's; renaming them forges + agreement the pin does not have; +- **passages quoting a live claim verbatim** (4) — `DDD-measure-12` and `term:verdict` still read + `ground distribution` at `v5.12.0`, so a rename would make the note misquote a live node; +- **one filename** (`measure-nonuniform-ground.py`, upstream) and **three occurrences ruled + ambiguous** between the population sense and the sense naming conditions in the case. + +**The pattern worth carrying:** a wave cannot move text it does not own. The three classes above are +all one rule — *the words are canon's* — and a wave's real scope is the prose its own artefacts +author. + +--- + +## 3. `measure-nonuniform-ground.py` and the section it computes now read under two words + +The measure note's §5.4 is renamed to *"`P` varied → a non-uniform deployment distribution"*. Its +reproducing asset is `core/assets/measure-nonuniform-ground.py`, upstream and out of a downstream +session's reach. The divergence is flagged in the note itself, in one clause, rather than hidden. + +**W2 closes it** when the migration reaches `core/assets/`. Until then the note names the asset by +its real name, which is the only option that keeps the reproduction instruction true. + +--- + +## 4. The measure note has no checkers + +Paper A has three — `check-quotations.py`, `check-appendix.py`, and now `check-status.py`. The +measure note has none, and its node table is hand-carried at `v5.7.0` while the paper's is generated +at `v5.12.0`. + +**The consequence is live, not theoretical.** This session's sixteen W1 deferrals in that note are +protected by one session's reading and by nothing else, and a later session renaming +`ground distribution` inside its `DDD-measure-12` quotations would break nothing any instrument +would notice. The note is also three releases behind on its own pin. + +The three scripts are already ref-parameterised and manuscript-parameterised, so pointing them at +the note is a small session rather than a new instrument. **It should happen before the note's pin +advances**, not after: the checkers are what make an advance observable. + +--- + +## 5. `README.md`'s "missing parameter" over-claims, and the README is the transfer's front door + +**Ruled at GATE 3 (Emil): goes to the freight list, repaired before publication rather than +whenever.** + +`README.md:32` reads: *"Decision-Driven Design is what you get when you fill in the actor slot those +results left empty — and discover that supplying the missing parameter changes their predictions."* + +That asserts an absence on the same grounds the paper's title asserted one, and the survey does not +support it: the arrangement as a unit of analysis is central to distributed cognition, joint +cognitive systems and systems-theoretic safety. The paper withdrew the claim and the title; the +README still makes it, in the first thing an outside reader sees. + +`CHANGELOG.md:303` uses the phrase too and is **deliberately left alone** — a changelog is a +historical record and is not edited retrospectively. + +--- + +## 6. An unverifiable attribution, recorded so it is not rediscovered and believed + +While drafting §11, a search result attributed a **"Law of Conservation of Complexity"** — +*"complexity is conserved under transformation and translation"* — to Woods & Hollnagel's *Joint +Cognitive Systems: Patterns* (2006). + +**If true it would be load-bearing.** It would give the framework's own conservation principle a +named precedent inside the literature the review says is unengaged, and it would change the Tesler +entry as well. + +**It could not be corroborated.** A curated index of the resilience-engineering literature's named +laws attributes the laws of fluency, stretched systems, coordinative entropy and the kludge to +Hoffman & Woods's *Beyond Simon's Slice*, and records no conservation law from *Patterns*. The +volume itself could not be obtained. **Nothing in the paper rests on it.** + +Anyone with the book in hand can settle it in a minute, and it is a real finding either way. diff --git a/papers/measure-note/measure-note.md b/papers/measure-note/measure-note.md index 0196ea3..8609d2e 100644 --- a/papers/measure-note/measure-note.md +++ b/papers/measure-note/measure-note.md @@ -24,7 +24,7 @@ of entropy: conditioning on any variable `X` splits the total into the verdict i `I(V;X)`, and what remains, `H(V|X)`. Three claims previously stated separately — seam demand under decomposition, store allocation across actors, and the encode/verify split — are this one identity under three choices of `X`. Two further instances, the identity iterated across a two-level -decomposition and a ground-distribution sweep, extend the worked coverage without adding a fourth +decomposition and a deployment-distribution sweep, extend the worked coverage without adding a fourth claim. **Demand is not cost.** Entropy prices what each act must resolve. Description length prices the @@ -38,8 +38,8 @@ correspondence that would make it a measured result — that `I(V;S)` predicts t interface — is stated as a protocol and not run. The measure has a sharp boundary, stated as a scope condition: the construction applies where the task -supplies an operationally usable verdict function and a ground distribution that can be estimated. -Outside that region the framework's independently derived floor becomes non-zero and the measure goes +supplies an operationally usable verdict function and a deployment distribution that can be +estimated. Outside that region the framework's independently derived floor becomes non-zero and the measure goes silent, for the same reason in both cases. That coincidence is worth noticing. It is not evidence, and the note does not use it as any. @@ -111,7 +111,7 @@ separates the two. The definition's phrase *the correct output the predicate ass task classes and not others; §2.2 says which, and what `V` is in the rest. **Notation.** `H(·)` denotes Shannon entropy in bits, `H(V) = −Σ_v P(v) log₂ P(v)`; `H(·|·)` -conditional entropy; `I(·;·)` mutual information. All are taken with respect to the ground +conditional entropy; `I(·;·)` mutual information. All are taken with respect to the deployment distribution `P`. Two counts appear and are never interchangeable: `n` is the number of points in the input space, used below as a display scale, and `N` is a number of acts (§6, §9). @@ -126,8 +126,8 @@ per-act claim. Three properties follow immediately, and each answers something the counting approach could not. -**It is representation-independent.** `H(V)` is a property of the verdict function and the ground -distribution. Describing the task at a different granularity does not change it. +**It is representation-independent.** `H(V)` is a property of the verdict function and the +deployment distribution. Describing the task at a different granularity does not change it. **It never mentions the actor.** This is the exact form of the framework's claim that demand is fixed by the task and never by the system. Whatever resolves the task faces the same `H(V)`. @@ -461,9 +461,13 @@ sub-decompositions; the arithmetic is identical, the instance distinct, and the remains unworked here [term:seam-identity; DDD-measure-14]. Iteration is arithmetic — the theorem is still Shannon's, applied twice. -### 5.4 `P` varied → non-uniform ground +### 5.4 `P` varied → a non-uniform deployment distribution -The worked example uses a uniform ground distribution. `P` is a parameter of the definition (§2), +*Worked by `measure-nonuniform-ground.py`, named before the vocabulary moved: the asset lives in +the principle repository and is out of this note's reach, so the section and the script that +computes it read under two words until the migration reaches `core/assets/`.* + +The worked example uses a uniform deployment distribution. `P` is a parameter of the definition (§2), not a convenience of the example, and the identity owes a demonstration under a skewed one [DDD-measure-12]. Re-running the task under two skewed deployments — benign, where valid inputs arrive nine times as often as invalid, and adversarial, the reverse — with both decompositions of §4: @@ -479,7 +483,7 @@ identity is indifferent to the skew; the demand is not. The same validator, unchanged, faces roughly four times the uniform demand when invalid inputs dominate and about a sixth of it when they are rare. Where the demand sits moves too: the share decomposition B carries in its seam is 41% of the whole on the benign -ground, 57% on the uniform, and 75% on the adversarial. A decomposition's seam economics are a +deployment distribution, 57% on the uniform, and 75% on the adversarial. A decomposition's seam economics are a property of the task *and its deployment*. *Fixed by the task* is fixed by the task, the tolerance, and the ground distribution — now worked, not merely stated. @@ -490,9 +494,9 @@ encode/verify split — are one identity under three choices of `X`. That a sing recovers all three, with no additional assumptions, is the note's main structural result. The two further instances add no fourth claim. Chaining exercises the same identity iterated; -the skewed ground exercises it with `P` varied. What they add is coverage: the identity has now been +the skewed distribution exercises it with `P` varied. What they add is coverage: the identity has now been worked on five instances — the three conditioning variables, a two-level chain, and a -ground-distribution sweep across three deployments — on two tasks. +deployment-distribution sweep across three deployments — on two tasks. **One caveat carried forward, and it is where the next work is.** In all three, *escape* is folded into `H(V|X)` together with *judgment*. The identity cleaves what `X` carries from everything else; it does @@ -598,7 +602,7 @@ cost layer downstream of this note [DDD-cost-06; DDD-cost-07]. This section does ### 6.5 Where demand comes due One consequence is worth stating and is not a further claim. Discharge is -**distribution-weighted**: demand comes due where acts concentrate, at the rate the ground +**distribution-weighted**: demand comes due where acts concentrate, at the rate the deployment distribution `P` supplies them. That is the measure's own `P` (§2) read back through the act rather than anything additional, which is why the aggregate above moves with `P` exactly as the per-act demand does (§5.4). The framework carries this in the same posture — as exposition, @@ -648,7 +652,7 @@ determination demand *is* verdict entropy. | It is **non-degenerate** — values are non-trivial, neither zero nor everything, and they move as the framework says they should | That the identification is the *correct* one | | The framework's qualitative claims appear with the **right signs and magnitudes** — a higher-information split really does buy cheaper parts; distractors really do push demand back to judgment | That information-theoretic demand predicts any **engineering** quantity | | The aggregate over a run is **exact and computable** — the correlation inequality's gap is §4's seam seen from repetition rather than decomposition (§6.3) | That correlated verdicts **are** cacheable work. That second identification is untested, exactly as the first is (§6.6) | -| No contradiction appears across five worked instances on two tasks — three conditioning variables, an iterated chain, and a three-deployment ground sweep | Anything whatsoever about open predicates | +| No contradiction appears across five worked instances on two tasks — three conditioning variables, an iterated chain, and a three-deployment distribution sweep | Anything whatsoever about open predicates | **So where does falsifiable content live?** In the identification, which is a modelling claim, and modelling claims fail by failure of correspondence rather than by arithmetic. The correspondence that @@ -710,7 +714,7 @@ never on live entropy. Necessary for the warrant, unnecessary for the operation The boundary is a scope condition, and it is best stated as one. > **The construction applies where the task supplies an operationally usable verdict function and a -> ground distribution that can be estimated.** +> deployment distribution that can be estimated.** Three requirements sit inside that sentence, and they fail in different ways. Collapsing them is the error the section is written to avoid. @@ -839,7 +843,7 @@ Where the framework has looked for this structure in production data, the evidence is consistent with the two-part form but cannot yet select the MDL form, and is filed as basis rather than confirmation [DDD-cost-03; mdl-cost-manufacturing-assessment-2026-08-08]. Within this note's own region a second, older -answer stands: entropy is relative to a declared ground distribution and computable from it +answer stands: entropy is relative to a declared deployment distribution and computable from it [DDD-measure-12], and Kolmogorov complexity is neither — which is what makes demand deployment-relative. The note therefore concedes the standing side to description length entirely: entropy cannot price the mechanism, and the cost layer built on this identity is MDL's. That concession @@ -896,7 +900,7 @@ actor-capacity model — stated as the next result and not worked here (§5.1, [DDD-cost-05; DDD-floor-01]. **Multi-actor composition remains unworked.** Of the instances previously owed, chained seams and -non-uniform ground are done (§5.3, §5.4); the composed-arrangement case is not +non-uniform deployment distributions are done (§5.3, §5.4); the composed-arrangement case is not [term:seam-identity]. Five worked instances is credibility, not certification, and the framing should be certified by an information theorist. The theorem is exact; identifying the right conditioning variable for a deployed system is estimation, with error bars. diff --git a/papers/paper-a/check-appendix.py b/papers/paper-a/check-appendix.py index bcbb471..9fb3a5f 100644 --- a/papers/paper-a/check-appendix.py +++ b/papers/paper-a/check-appendix.py @@ -12,9 +12,14 @@ 3. every row corresponds to a node the body cites (nothing invented); 4. the hypothesis-set rows really do carry empty evidence. -Usage: check-appendix.py +Usage: check-appendix.py [] Exit 0 when the appendix matches the graph; 1 otherwise. +Citations are read from ; the rendered appendix is read from +when the two are separate files. Checks 2 and 3 -- nothing dropped, nothing invented -- are +exactly the checks that keep a split appendix honest, so they run ACROSS the two files rather +than within one. + Defect history, kept with the instrument because a check that can be wrong about a correct artefact could have been wrong about an incorrect one: @@ -72,17 +77,21 @@ def strip_quote(md): return flat(' '.join(re.sub(r'^> ?', '', ln) for ln in str(md).split('\n'))) -def main(path, repo, ref): +def main(path, repo, ref, home=None): nodes = graph(repo, ref) # Which ids are claims, so a decisions row (id | statement) is not mistaken for a claims # row that lost its metadata columns. claim_ids = {i for i, n in nodes.items() if 'status' in n and not i.startswith('term:')} - text = open(path).read() - head, _, appendix = text.partition('## Appendix A') + head = open(path).read().partition('## Appendix A')[0] + _, _, appendix = open(home or path).read().partition('## Appendix A') if not appendix: - sys.exit('no Appendix A found') + sys.exit(f'no Appendix A found in {home or path}') body_cited = set(re.findall(r'DDD-[a-z]+-\d+', head)) | set(re.findall(r'term:[a-z-]+', head)) + if not body_cited: + sys.exit(f'{path} cites no graph ids -- the manuscript is the first argument and the ' + 'appendix home the fourth; passing the appendix as the manuscript would report ' + 'every row as uncited') rendered, bad = set(), [] # Statements carry literal pipes -- H(V|X), H(V|S) -- escaped as \| in a cell. Splitting on @@ -157,6 +166,6 @@ def main(path, repo, ref): if __name__ == '__main__': - if len(sys.argv) != 4: + if not 4 <= len(sys.argv) <= 5: sys.exit(__doc__) sys.exit(main(*sys.argv[1:])) diff --git a/meta/sessions/2026-08-30-paper-a-revision/status-sweep.py b/papers/paper-a/check-status.py similarity index 51% rename from meta/sessions/2026-08-30-paper-a-revision/status-sweep.py rename to papers/paper-a/check-status.py index 71f7f77..4805d1b 100755 --- a/meta/sessions/2026-08-30-paper-a-revision/status-sweep.py +++ b/papers/paper-a/check-status.py @@ -1,18 +1,44 @@ #!/usr/bin/env python3 -"""Inline status assertions in a projection, checked against the graph at a ref. - -The gap this closes. `check-quotations.py` verifies BLOCK quotations, and its docstring says so -honestly. But the manuscript's more common form is the inline citation that asserts a status in -prose -- `[DDD-measure-06, **established**]` -- and no instrument reads it. At the v5.9.0 -> v5.12.0 -advance that gap hides a retirement: DDD-measure-06 is retired at v5.12.0, the paper calls it -established three times, and both existing checkers pass those three lines. - -A status label is the part of a citation a reader converts into warrant. It is therefore the part -most worth checking, and it was the part with no check. - -Usage: status-sweep.py [] +"""Verify every inline status assertion in a projection against the claim graph at a ref. + +The gap this closes, and it is a third gap rather than a variation on the other two. +`check-quotations.py` verifies BLOCK quotations and says so honestly in its own docstring; +`check-appendix.py` re-reads the generated node tables. Neither reads the manuscript's MORE COMMON +form: the inline citation that asserts a status in running prose -- `[DDD-measure-06, +**established**]`. At the v5.9.0 -> v5.12.0 advance that gap hid a retirement. DDD-measure-06 is +retired at v5.12.0; the paper called it **established** in three places; both existing checkers +passed all three lines, because none of the three is a block quotation and none is an appendix row. + +**A status label is the part of a citation a reader converts into warrant.** It is therefore the +part most worth checking, and it was the part with no check. That the review's central criticism was +that the paper communicates more warrant than the graph provides, and that the one unchecked +surface was exactly the warrant labels, is not a coincidence worth leaving unfixed. + +The script also reports every cited node whose status MOVED between two refs, label or not, when a +baseline ref is given. A status can move under a citation that never named one while the prose +around it still asserts the old reading -- which is what happened at §8.4 and in the conclusion's +"what is established" list, neither of which carried a label. + +Usage: check-status.py [] Exit 0 when every inline status assertion matches the graph at ; 1 otherwise. -Appendix A is excluded: it is generated wholesale and re-read by check-appendix.py. +Appendix A is excluded, wherever it lives: it is generated wholesale and re-read by +check-appendix.py, and checking it here would report one defect twice. + +Defect history, kept with the instrument: + + * Written as a one-off sweep at this revision's GATE 1 to answer a question -- does the pin + advance break anything the two checkers cannot see -- and it did, four times. It was kept as + a session artefact for one gate before being ruled an instrument. Recording that here because + the instrument's origin is its warrant: it was not designed against a specification, it was + generalised from a sweep that found something, and its coverage is therefore exactly the + coverage of that sweep and no wider. + + * The citation regex accepts a status after a comma or an em dash, matching both forms the + manuscript uses -- `[id, projected]` and `[id -- closing clause]`. A citation carrying NO + status word is skipped rather than flagged: a projection may cite without asserting, and + treating a bare citation as a defect would make the honest form the expensive one. The cost + is that a bare citation whose surrounding PROSE asserts a status is invisible here, and the + baseline-ref comparison is the mitigation rather than a cure. """ import re import subprocess diff --git a/papers/paper-a/gen-appendix.py b/papers/paper-a/gen-appendix.py index c837a32..1aff616 100644 --- a/papers/paper-a/gen-appendix.py +++ b/papers/paper-a/gen-appendix.py @@ -7,9 +7,14 @@ result is then re-read against the graph by a separate script, `check-appendix.py`, so the check is independent of the generator that produced it. -Usage: gen-appendix.py +Usage: gen-appendix.py [] Rewrites everything from the `## Appendix A` heading to end of file. +Citations are always read from . When is given, the appendix +is rendered THERE and the manuscript is not written at all -- which is how the supplement split +works: the argument stays in the paper, the generated graph tables live beside it, and the two +files cannot drift because one is still generated from the other's citations. + Defect history, kept with the instrument: * First version was not idempotent: it appended a horizontal rule to a body that already @@ -18,6 +23,14 @@ did not. Fixed by stripping trailing rules from the body first; idempotence is now checked by running three times and comparing bytes. + * The appendix's home became a second file at the supplement split, and the first attempt + passed the supplement as the manuscript. Citations are then read from the supplement, which + cites nothing, so the generator rendered an EMPTY appendix and exited 0. A generator whose + correct output for a correct input is indistinguishable from its output for the wrong input + is not a generator you can run unattended. The two roles are now separate arguments and the + citation source is never the render target unless they are the same file by intent; a render + that finds zero cited ids exits non-zero rather than writing an empty table. + * The claims table rendered `id | status | statement` and no kind, for five minor versions. `kind` has been populated on every claim since format 1, so this was never a schema gap -- an external reader inferred one from the appendix and reported the registry as missing a @@ -104,13 +117,16 @@ def canonical(term): return cell(' '.join(lines)) -def main(path, repo, ref): +def main(path, repo, ref, home=None): claims, decisions, terms = load(repo, ref) text = open(path).read() body = text.split(HEADING)[0].split('## Appendix A')[0] cited_c = sorted(set(re.findall(r'DDD-[a-z]+-\d+', body))) cited_t = sorted(set(re.findall(r'term:[a-z-]+', body))) + if not cited_c and not cited_t: + sys.exit(f'{path} cites no graph ids -- refusing to render an empty appendix. ' + 'Citations are read from the manuscript; the appendix home is the 4th argument.') miss = [c for c in cited_c if c not in claims and c not in decisions] miss += [t for t in cited_t if t not in terms] if miss: @@ -154,16 +170,23 @@ def main(path, repo, ref): # The body already ends with the rule that preceded the appendix. Appending another would # make the generator non-idempotent -- a second run would add a second rule, and a convention # that says "regenerate wholesale" must survive being run twice. - body = body.rstrip('\n') - while body.endswith('---'): - body = body[:-3].rstrip('\n') - open(path, 'w').write(body + '\n\n---\n\n' + '\n'.join(out)) + if home is None or home == path: + target, keep = path, body + else: + # Rendering into a separate home: the manuscript is not written at all, and whatever the + # home carries before the heading -- its own front matter -- is preserved verbatim. + target = home + keep = open(home).read().split(HEADING)[0].split('## Appendix A')[0] + keep = keep.rstrip('\n') + while keep.endswith('---'): + keep = keep[:-3].rstrip('\n') + open(target, 'w').write(keep + '\n\n---\n\n' + '\n'.join(out)) print(f'Appendix A generated: {len([c for c in cited_c if c in claims])} claims, ' f'{len([c for c in cited_c if c in decisions])} decisions, {len(cited_t)} terms, ' f'{len(hset)} hypothesis-set rows') if __name__ == '__main__': - if len(sys.argv) != 4: + if not 4 <= len(sys.argv) <= 5: sys.exit(__doc__) main(*sys.argv[1:]) diff --git a/papers/paper-a/paper-a-supplement.md b/papers/paper-a/paper-a-supplement.md new file mode 100644 index 0000000..e5649c8 --- /dev/null +++ b/papers/paper-a/paper-a-supplement.md @@ -0,0 +1,268 @@ +# Supplement — *Actor-Indexed Determination* + +*Companion to `paper-a.md`. Not required to follow the argument.* + +--- + +## S1. What this supplement is, and why it is separate + +An external review found the paper carrying its own filing history, pending-node discussion and +generated graph tables inside the argumentative path, and judged that an archival paper should not. +The finding was accepted. **The apparatus is not reduced; it is moved.** Everything here still runs, +still verifies, and still fails loudly. + +The boundary applied is one rule: + +> **A passage moves here when it is about the graph. It stays in the paper when it is about the +> world, or about how much warrant the paper has.** + +So the status *vocabulary* is here and the sentence *"this claim is `projected`, its evidence field +is empty, and the study is unrun"* is not — that one is about warrant and stays where a reader meets +the claim. Filing history, minting rulings, pin mechanics and reproduced node tables are about the +graph. + +--- + +## S2. The projection's pin, and what it is pinned to + +The paper is a projection of `actor-indexed-determination` at the annotated tag **`v5.12.0`**. Every +bracketed claim, decision and term identifier in the paper resolves against that one ref. No +identifier resolves against the downstream repository, so no downstream ref is pinned. + +**The pin was held deliberately, and then advanced.** The paper was merged as a projection of +`v5.9.0`, and the pin was kept there by ruling while canon moved — through `v5.10.0`, `v5.11.0` and +`v5.12.0` — so that upstream repair would not drag the projection behind it. This revision advances +it in one step. Two of the four quotations that broke at the advance are quotations of claims the +review asked the paper to stop overstating, so the repair paid a debt the paper already owed. + +The companion measure note — `papers/measure-note/measure-note.md` in `decision-driven-design` — is +cited by path at commit `d6bc1a1`, which carries its discharge section and the vocabulary rename +described in S5. The note is absent from `v0.4.0` and from every earlier downstream tag, so it +resolves at a commit rather than at a tag. + +--- + +## S3. Filing status of what the paper carries + +Three passages in the paper state a distinction the graph does not carry. Each is marked +register-native at the point of use; this is where their filing status is recorded. + +| Passage | Status in canon | Why | +|---|---|---| +| §2.4's five-way ground provenance (controlled, observed, inferred, institutional, missing) | **ruled ineligible for minting**, not merely deferred [DDD-dec-26] | a mint would fix a partition that an open trust question may restructure, and the institutional slot's mechanism is gated on it. §8.1 exhibits a case the partition does not adjudicate, which is the reason arriving in a worked form | +| §5.2's **constructively closed** rung | **not filed**; no claim node names it, and the word *constructive* occurs nowhere in the principle repository's core documents | `term:closure` is stated in evaluative terms alone. A dedicated node is pending on the framework's open-questions wave; until it lands the rung's basis is closure, the separation of closure from generation cost, and the measure's silence on search | +| §3's two commitment terms | **`draft`** — filed, not ratified [DDD-dec-26] | both were minted only when a downstream projection needed them as citable nodes, and a downstream document is never the proximate cause of an upstream mint: the canonical text was drawn from canon's own existing exposition | + +**The paper files nothing.** A projection that mints is a projection that has stopped being one. + +--- + +## S4. A finding this revision leaves open in canon + +The paper's §7 reports that the graph carries **two counts** of accountability's elements — +`DDD-frame-08`'s five and `term:accountability`'s three — with a mapping showing that the five +refines the third element into stake and sanction path and **adds** authority linkage. + +A `projected` claim that adds an element to a `settled` term is a supersession question, not an +elaboration. **The paper flags it and does not take the ruling**, which is not a projection's to +take. The finding is filed as a successor item against both nodes, naming this session as its +finder, so that the flag is delivered to canon rather than sitting in a projection nothing points +back at — a flag nothing points back at being, by this framework's own vocabulary, filed and +undelivered. + +--- + +## S5. Vocabulary: `ground` in its population sense + +This revision carries the first wave of a vocabulary migration. The word `ground` had been carrying +five senses across the corpus, and the **population sense** — the distribution of cases a task +faces — is renamed **`deployment distribution`** throughout this paper and the measure note, where +those two artefacts own the prose. + +Three kinds of occurrence were deliberately **not** renamed, and each for a reason that is the same +reason: **the words are canon's, not the paper's.** A reproduced node table carries the graph's own +statements; a passage quoting a live claim verbatim must keep quoting it; and canon still reads +`ground distribution` at `v5.12.0`. Renaming any of them would forge agreement the pin does not +have, which is the same principle the pin itself exists to enforce. Those occurrences move when +canon moves. + +Three further occurrences were ruled **ambiguous** between the population sense and the sense +naming the conditions in the case, and deferred whole rather than ruled in a paper session. Where a +rename is contestable, it is not this wave's. + +--- + +## Reproduction + +**This paper mints no figures and no assets.** Every number it states was produced by an existing +script in the principle repository and is cited to the projection that works it: the +date-validation totals of §4.4 and §8.6 come from `core/assets/measure-toy.py`, worked in the +companion measure note's §4 and in `core/09` §3. That script was re-run fresh while this paper was +drafted and reproduces the stated values. + +**Three checkers, and each covers a surface the others cannot reach.** All three live beside the +manuscript and take the ref as an argument, so every check is reproducible against any ref rather +than against the one that happened to be current. + +| Script | What it verifies | Surface | +|---|---|---| +| `check-quotations.py` | every block quotation attributed to a node is verbatim at the ref, or discloses itself partial in its own citation | block quotations | +| `check-appendix.py` | every rendered row matches the graph; nothing the paper cites is missing; nothing is invented; the hypothesis rows really do carry empty evidence | the generated tables | +| `check-status.py` | every inline citation asserting a status asserts the status the graph carries | running prose | + +The third is new at this revision, and its origin is a defect rather than a design. The pin advance +retired `DDD-measure-06`, the paper called it **established** in three places, and the two existing +checkers passed all three lines — because none was a block quotation and none was an appendix row. +The unchecked surface was precisely the one that converts a citation into warrant. + +**Appendix A is generated wholesale and never hand-edited** (`gen-appendix.py`), then re-read +against the graph by an independent script that shares no code with the generator +(`check-appendix.py`). Since the split, the generator reads the manuscript's citations and renders +into this file, so the two cannot drift: the supplement's tables are a function of what the paper +cites. Idempotence is checked by running the generator three times and comparing bytes. + +Reproducing the whole apparatus, from the directory holding the manuscript: + +``` +python3 check-quotations.py paper-a.md v5.12.0 +python3 check-status.py paper-a.md v5.12.0 +python3 gen-appendix.py paper-a.md v5.12.0 paper-a-supplement.md +python3 check-appendix.py paper-a.md v5.12.0 paper-a-supplement.md +``` + +**What the checkers do not do.** They verify correspondence between this paper and the graph. They +do not establish that anything in the graph is true, and a fully green run is consistent with every +claim in the paper being wrong. + +--- + +## Appendix A. Cited claims, decisions and terms + +The paper cites nodes in the framework's claim graph. Statements below are reproduced +word-for-word from the graph at the ref pinned in the front matter, so the paper can be checked +without it. **Kind** and **status** are the graph's own fields, and they answer different questions. +*Kind* is what sort of claim it is: *formal* is arithmetic or a derivation, *empirical* rests on +observation, *conceptual* fixes or uses the framework's vocabulary, *normative* says what ought to +be done. *Status* is how far it has been argued: *settled* and *established* are argued and +unchallenged **within this framework**, *reported* is exercised by a reproducing computation, +*projected* is proposed with a declared falsifier and not yet met, *draft* is filed and not yet +ratified, *retired* is superseded and kept with the correction that killed it. + +**Neither field claims external validation, and the two must be read together.** *Established* +means internally argued and unchallenged, not empirically confirmed; *reported* means a computation +runs and reproduces, not that the world was consulted. The pairing is what carries the information: +every *established* claim in this graph is *formal*, so what is settled here is arithmetic, and the +modelling claims that give the arithmetic its meaning are *projected*. The canonical statement of +what each value means, and what it does not, is `spec/claim-format.md` §5 at the pinned ref; this +paragraph projects it and does not replace it. + +**This appendix is generated from the graph and never hand-edited** (`gen-appendix.py`), then +re-read against the graph by an independent script (`check-appendix.py`). + +### Claims + +| ID | Kind | Status | Statement | +|---|---|---|---| +| `DDD-agent-01` | empirical | projected | Long-running agent drift is escaped decisions caused by basis loss: context decay, compaction, and distractors remove claim nodes from the agent's ground, so later actions are decisions with no basedOn edge to the declared claims and revert to model priors. Grounding the agent in a persistent external claim graph — basis as query, not context residue, with per-decision claim citation — reduces drift relative to context-carried instruction. | +| `DDD-cost-08` | conceptual | projected | Actor selection for an act is two-gated: capacity gates always — the actor must carry the act's residual at the declared tolerance or the excess escapes — and assurance gates exactly where the acceptance predicate does not close, where assurance must attach to the actor because no check can carry it. | +| `DDD-cost-09` | conceptual | projected | Assurance-by-actor binds assurance to a scarce carrier, supplied occasioned — at the act, at the carrier's class price. Assurance-by-check moves it into a mechanism, supplied standing — independent of the act. Closing a predicate converts a property's assurance supply from occasioned to standing. | +| `DDD-cost-11` | conceptual | projected | On an open predicate, assurance and actor class are positively coupled; closing the predicate flips the sign — the assurance gate lifts and the capacity gate softens — so the actor class the act requires falls, leaving the actor carrying generation only. | +| `DDD-cost-12` | conceptual | projected | Required actor class for an act is the maximum, over the act's capabilities, of the class needed where assurance is not mechanically discharged — per capability, not per act. | +| `DDD-cost-13` | conceptual | projected | Where assurance attaches to the actor per capability, an answer-keyed qualification instrument — an examination, an eval — evidences demonstrated class on predicates that close. Its verdict cannot evidence open-predicate carriage: the instrument's own predicate closes while the target predicate does not, so delegation to the open predicate substitutes the actor's identity for exactly the check the instrument cannot be. | +| `DDD-cost-20` | conceptual | projected | Encoding around a carrier and encoding within it differ in locus: around-encoding — context, retrieval, scaffolds — is standing supply outside the carrier, delivered through the channel at each act; within-encoding — training — converts judgment allocation to encoded allocation inside the carrier. Training buys allocation, not capacity: it does not enlarge the judgment store. | +| `DDD-cost-25` | conceptual | projected | Assurance mechanisms occupy temporal positions relative to the act — pre-act (selection, training, encoding, static checks), at-act (monitoring), post-act (review, audit, consequence) — each with a latency; a mechanism whose latency exceeds its position's budget — the episode for at-act mechanisms, the consequence horizon for post-act ones — cannot hold its position, so rising tempo, which compresses both budgets, forces assurance pre-act, into standing supply or the carrier. | +| `DDD-delivery-01` | conceptual | projected | Filing is not encoding: a decision sits in I(V;X) only to the extent the arrangement delivers it at the act, so store allocation cannot be read off artefacts, and the paid-once-inherited-by-every-run property belongs to act-triggered delivery specifically, not to standing supply generally. | +| `DDD-delivery-02` | conceptual | projected | Governance filed but not delivered is escape: no source supplied the governing decision at the act, so it was determined by nobody (term:escape, supply-general), and delivery failure is thereby a generator of escape — sufficient, never necessary — whose distinguishing feature is that the ledger shows coverage: escape that presents as governance. | +| `DDD-delivery-03` | conceptual | projected | An unretrieved decision and an unretrieved check over the same act are correlated failures — same actor, same budget, same position — so judgement-mediated delivery on both the source and assurance sides silently removes the independence a gate depends on: the failures compound rather than stack. | +| `DDD-floor-01` | formal | reported | H(V\|X) bundles judged and escaped demand; cleaving them requires an actor-capacity model, and residual demand an actor has taken up escapes where it exceeds effective capacity min(C_hold, C_resolve) AND the shed decisions carry no verifier — overflow ∩ open is the mechanism of capacity-generated escape, sufficient for escape and not necessary for it, with overflow alone producing retries, not escape. | +| `DDD-floor-02` | conceptual | projected | The judgment floor is relational: irreducibility is a property of the indexed relation ⟨task, ground, acceptance relation, tolerance, arrangement, assurance⟩, not of the task alone — the portion of determination an arrangement cannot discharge through its prior commitments or adequate direct verification at the declared assurance level moves when any coordinate of the relation moves. | +| `DDD-frame-01` | conceptual | projected | Unresolved determination is indexed by the tuple ⟨task, ground, acceptance relation, tolerance, arrangement, assurance⟩, not by the task alone. | +| `DDD-frame-02` | conceptual | projected | Behavioural commitments attach at three levels — outcome, policy, principal — which compose and are not actor species; residual discretion is the outcome-relevant variation those commitments leave open at the act, held at fixed ground. | +| `DDD-frame-03` | conceptual | projected | The source of a resolution and the mechanism assuring it are separate dimensions; the four-store model held the arrangement largely fixed while the allocation was analysed, and did not draw the distinction. | +| `DDD-frame-04` | empirical | projected | Escaped decisions — consequential resolutions with no adequate source-and-assurance combination — predict ungoverned failure modes and design-review findings. | +| `DDD-frame-05` | formal | projected | Under a sound terminating operational checker with complete declared ground, producer identity is not epistemically necessary for the checked property — and nothing more: not cheap generation, not normative completeness, not accountability. | +| `DDD-frame-06` | formal | established | Closure is distinct from generation cost: verification being cheap implies nothing about the density or accessibility of the acceptance region. | +| `DDD-frame-07` | empirical | projected | Operational evaluability, feedback density, and ground accessibility predict the comparative advantage of computationally assisted arrangements over unaided situated judgment, after controlling for difficulty and resources (H1–H5, filed as DDD-hyp-01 through DDD-hyp-05). | +| `DDD-frame-08` | normative | projected | Accountability is a relation (attribution, persistent principal, authority linkage, stake, sanction path), not an intrinsic capacity; an arrangement naming an executor but no principal is incomplete. | +| `DDD-frame-09` | formal | retired | RETIRED — "closed predicates make intelligence unnecessary." Does not follow from producer-independence under verification; generation may still require whatever capability the search demands. | +| `DDD-frame-10` | formal | projected | Determination demand is conserved as a scalar across arbitrary re-arrangements, actor-generally and including open predicates. | +| `DDD-frame-11` | conceptual | projected | The governance question — is every decision governing the act in a declared store, none escaped? — is well-formed on the total domain, including open predicates where the measure does not exist; the cost question — how much is in each store? — exists only where the predicate closes. The framework's governed domain is strictly wider than its measured domain. | +| `DDD-frame-13` | conceptual | projected | Determination's object is the determinable — an outcome-relevant dimension of variation at the declared tolerance — and its product is a determinate, one specific way of occupying that dimension; determination demand is thereby a measure over unresolved determinables, the verdict variable's support being the determinate-space at the declared grain where the predicate closes. | +| `DDD-frame-14` | conceptual | projected | Discharge always produces a determinate, which lands in two registers: as an outcome — the determinate as it lands in the world, produced at every completed act — and as a verdict — the determinate as assessed by a declared predicate, produced only where governance has declared one; governance is thereby the conversion of outcomes into verdicts, and every diachronic instrument runs on verdicts alone. | +| `DDD-frame-15` | conceptual | retired | RETIRED — "every completed act's determination demand is discharged by a filed decision, an actor's judgment, an arrangement default, or an uncontrolled draw." The four modes do not partition: a declared default is both a filed decision and an arrangement default, a thermostat satisfies the judgment gloss under a rule canon times as encoded, and trained inference is policy commitment, judgment and standing encoded supply at once. Superseded by DDD-frame-17, which partitions the same object by dichotomy on determination and control. | +| `DDD-frame-16` | conceptual | projected | Discharge is act-indexed: standing supply is inherited per act and occasioned supply is produced per act, so there is no act-free discharge — governance never chooses whether demand is supplied, only by what, chosen in advance or defaulted at the act. | +| `DDD-frame-17` | conceptual | projected | At every completed act in a task's scope, each outcome-relevant alternative is discharged in exactly one of three ways: predetermined — the arrangement's standing configuration together with the ground at the act determines the resolution; exercised — it does not, and something within the arrangement's control determines it at the act; or drawn — it does not, and what determines it lies outside the arrangement's control. Demand is never unmet, only ungoverned. | +| `DDD-ground-01` | normative | projected | A governing decision must declare a resolvable applicability predicate, unless it explicitly declares universal applicability; non-evaluation must never silently become non-applicability, and where the predicate is implemented over declared ground axes, each axis is marked mechanically-evaluable or judgement-evaluable. | +| `DDD-ground-02` | conceptual | projected | Source coverage (covered · declared-empty · undeclared · unknown), resolution (resolved · deliberately-open · unknown), and assurance (adequate · inadequate · unknown) are orthogonal properties of ground relative to a filed decision set: only source coverage = undeclared is a source-coverage finding, deliberately-open is a resolution value carrying a deferred verdict, and Unknown is never a pass. | +| `DDD-ground-03` | conceptual | projected | A decision whose resolution is deliberately-open has no resolution for a timing predicate to read, so the timing vocabulary carries a fourth value — "—(open)" — alongside before, during, and after; any timing predicate without it misfiles open decisions, because every definite value it can assign erases the declaredness that makes them open. | +| `DDD-ground-05` | conceptual | projected | Declaring the determinable space is constitutively prior to determination over it — a determination selects a determinate, and determinates exist only as ways of occupying a declared determinable — with the symmetry that ground is prior within each act while decisions are prior in a registry's growth, so the priority is synchronic constitution and the bootstrap diachronic history, not a circle. | +| `DDD-hyp-01` | empirical | projected | Holding generation difficulty and resources constant, comparative advantage shifts toward high-throughput computational generators as acceptance becomes more operationally evaluable, feedback becomes faster and denser, ground becomes more accessible, checking becomes cheaper, and retries become more affordable. | +| `DDD-hyp-02` | empirical | projected | Human or institutionally situated arrangements retain greater comparative advantage as relevant ground becomes unavailable to the computational system, consequences are delayed, evaluators disagree, acceptance criteria drift over time, tacit or socially distributed knowledge is required, or normative legitimacy is part of the task. | +| `DDD-hyp-03` | empirical | projected | A generator-plus-checker arrangement — a computational generator composed with a mechanical checker or an independent reviewer — outperforms both generator-alone and judgment-alone baselines where candidate generation benefits from breadth or speed, significant parts of acceptance are operationally closed, and the remaining open residue can be escalated. | +| `DDD-hyp-04` | empirical | projected | Trust and deployment willingness are better predicted by the completeness of the accountability arrangement — attribution, persistent principal, authority linkage, stake, sanction path — than by whether the immediate executor is human or computational. | +| `DDD-hyp-05` | empirical | projected | Reliance on worker or provider selection increases as result-level evaluation becomes slower, less objective, less stationary, and less complete, holding labour supply, training cost, consequence severity, and task structure constant. | +| `DDD-measure-01` | empirical | projected | Specification demand is verdict entropy: for a task whose acceptance predicate closes, the demand engineers experience as specification burden is H(V) over the ground distribution. | +| `DDD-measure-02` | formal | established | Given the identification (DDD-measure-01), conservation on the closing region is the chain rule of entropy: H(V) = I(V;X) + H(V\|X) for any conditioning variable X. | +| `DDD-measure-03` | formal | reported | The seam of a decomposition is I(V;S); a decomposition with cheaper parts has pre-paid more demand into the seam, and H(V\|S) is minimised exactly when I(V;S) is maximised. | +| `DDD-measure-06` | formal | retired | RETIRED — "the measure exists iff the acceptance predicate operationally closes; H(V) is undefined exactly where the framework's floor result locates non-zero floor." The biconditional fails in both directions and the two limbs carry different warrants under one status. Superseded by DDD-measure-16 (availability to an arrangement, with existence and estimability separated) and DDD-measure-17 (the coincidence, at projected). | +| `DDD-measure-08` | formal | retired | RETIRED — "a better decomposition destroys demand." Cheaper parts were purchased by a higher-information seam; the destruction was an artifact of not counting I(V;S). | +| `DDD-measure-10` | formal | established | You cannot decompose your way out of the work: for a fixed closing task, H(V\|S) = 0 requires I(V;S) = H(V) — the parts become trivial only when the decomposition already encodes the entire verdict. Demand is conserved, not escapable by re-decomposition. | +| `DDD-measure-11` | conceptual | reported | The measure prices the verdict, not the search: H(verdict) is a property of the verdict function and the ground distribution and says nothing about the cost of computing a correct answer. Two tasks with identical verdict entropy can differ unboundedly in generation cost, so the measure must not be read as pricing generation. | +| `DDD-measure-15` | conceptual | projected | The engineering reading of the chain-rule identification holds only for admissible conditioning variables. A conditioning variable X is admissible where it is computable from ground available at the act and from what the arrangement has standing before it, and not from the verdict itself — computable by something that has not been handed the answer. The arithmetic holds for any X whatever; admissibility is what restricts the reading, not the identity. | +| `DDD-measure-16` | formal | established | The verdict-entropy construction is available to an arrangement exactly where the task's acceptance predicate closes for that arrangement; availability is a property of the arrangement, not of the task, and it is neither the existence of the verdict function nor the estimability of the ground distribution. | +| `DDD-measure-17` | conceptual | projected | The verdict-entropy construction's domain and the region where the framework's floor result locates non-zero floor coincide, and the coincidence is principled rather than evidential: the two arguments share the closure of the acceptance predicate as a premise, so their agreement about where the line falls is close to definitional on the measure's side and is not evidence that the identification is correct. | + +### Decisions + +| ID | Statement | +|---|---| +| `DDD-dec-15` | The escape mechanism's scope correction: overflow ∩ open is the mechanism of capacity-generated escape — sufficient for escape, never necessary for it. Escape as such stays as term:escape defines it, supplied by nobody for any reason, and term:escape-mechanism and DDD-floor-01 are re-scoped by supersession of their universal quantifier. | +| `DDD-dec-24` | Wave 3 files the principle layer's central material: the indexed-determination core given its canon home (core/14, with the flag clearances that home licenses), the discharge chain (the determinable, the determinate's two registers, supply-mode exhaustiveness, act-indexed discharge, constitutive priority of ground), and the hypothesis set as a new claim area — six derivation-grade claims, five statement-grade hypotheses, three settled terms, one new core document, filed by supersession-and-amendment with every divergence from source material reported rather than harmonised. | +| `DDD-dec-26` | The Track 1 session mints two terms and no more — term:commitment-level and term:residual-discretion, both established by core/14-indexed-determination.md §2 — discharging the mint DDD-frame-02 deferred pending use; three anticipated deltas (axis, arrangement, escaped decision) are ruled non-deltas because canon already carries each, and the five-way ground-provenance taxonomy is ruled ineligible for minting because its institutional slot is Q27-gated. | + +### Terms + +| ID | Term | Canonical wording | +|---|---|---| +| `term:accountability` | accountability | **Accountability** is a property of the arrangement, not of the executor: attribution of the determination, a persistent answerable party, and a borne consequence. An arrangement missing any of the three has not allocated the decision's consequence. | +| `term:actor` | actor | An **actor** is a system that resolves decisions by reading ground: variation in declared ground can alter the resolution through an internal pathway that selects among alternatives. A thermostat qualifies; a falling rock does not. Actorhood does not require intelligence. | +| `term:admission-test` | admission-test | **A choice is a decision iff varying *the choice* moves the outcome past tolerance.** **A fact is ground iff varying *the world* moves the outcome past tolerance.** | +| `term:answerability` | answerability | **Answerability** — the obligation to produce the chain: which determinations were made, by whom, against what ground. | +| `term:arrangement` | arrangement | The **arrangement** is the composition through which a resolution is produced and governed: executor, prior commitments, ground channels, checks, reviewers, record, and accountable principal. The unit of comparison is the arrangement, not the isolated actor. | +| `term:attribution` | attribution | **Attribution** — provenance-shaped, and therefore checkable: the record connecting the determination to the execution that produced it. | +| `term:closure` | closure | **Effective closure, defined.** A predicate is **closed for an arrangement** when the relevant ground is observable and adequacy can be evaluated within declared resource, latency, and confidence bounds. **Decidable** is reserved for the formal special case. | +| `term:commitment-level` | commitment-level | A **commitment level** is a level at which an arrangement fixes behaviour in advance: **outcome-level** — permitted resolutions fixed directly; **policy-level** — the generating procedure fixed; **principal-level** — a determiner selected by qualification and case-level resolution delegated. The three compose, and they are levels of commitment, not species of actor: the question is never which of three kinds an actor is, but at which levels the arrangement has committed. | +| `term:composite-actor` | composite-actor | A **composite actor** — several actors read as one at a declared boundary: from outside it the composite is **one actor**, with one capability envelope and one verdict owed. **A composite actor carries its members' demand, plus the seam demand between them.** | +| `term:conservation` | conservation | **For a task at a declared assurance level, and within a fixed decomposition of that task, determination demand is conserved.** Every governing decision gets made. The only choice is **by whom, when, and at what price.** Reduce the demand in one store and it **relocates**; it does not vanish. | +| `term:determinable` | determinable | The **determinable** — an outcome-relevant dimension of variation at the declared tolerance: the object determination resolves, and the dimension of comparability an axis names. Determinateness comes in orders — red → scarlet → this shade — and the declared tolerance names the order at which the framework stops distinguishing. | +| `term:determinate` | determinate | The **determinate** — one specific way of occupying a determinable: what discharge produces. A determinate is a way of being, not the determinable plus a differentia, and determinates under one determinable are constitutively exclusive at their grain. | +| `term:ensemble-actor` | ensemble-actor | **The choice is a property of the ensemble, and it exists nowhere in any member.** | +| `term:escape` | escape | **Escaped** — determined *never*, by nobody: decided-by-nobody as a first-class category. Latent defect exposure. **The only forbidden state.** | +| `term:escape-mechanism` | escape-mechanism | **Capacity-generated escape — the escape an actor produces from residual it has taken up — requires two conditions, both necessary:** **(1) Overflow** — demand exceeds resolve capacity. **(2) Open** — no verifier the actor holds. Overflow alone (closing predicate) → **retries, not escape.** Recoverable. Not floor. Open alone (within capacity) → **carried by judgment**, where an accountable supplier is named. Not floor. Where none is named, it is escape by another route (`05` §7) — outside this mechanism, not excluded by it. **Overflow AND open** → **escape. This is the floor.** **Sufficient for escape, never necessary.** A governing decision that never entered an actor's residual escapes without overflowing anything: escape is supplied-by-nobody (`term:escape`), and capacity shortfall is one generator of it. | +| `term:floor` | floor | The "floor" is the portion of a determination's demand that **cannot be moved off the in-the-moment actor** — the residue that no amount of encoding or checking can amortise, that must be paid, per run, in judgment. **The intrinsic floor is a property of the acceptance predicate, not of the decision.** | +| `term:liability` | liability | **Liability** — bearing the consequence. | +| `term:outcome` | outcome | The **outcome** — the determinate as it lands in the world, produced at every completed act. The **verdict** is the same determinate as assessed by a declared predicate, produced only where governance has declared one. The world renders outcomes, never verdicts; governance is the conversion of outcomes into verdicts. | +| `term:residual-discretion` | residual-discretion | **Residual discretion** is the outcome-relevant variation the arrangement's commitments leave open at the act, **held at fixed ground**: the alternatives still admissible once the standing configuration and the ground at the act are both given. It is not variation across ground — a cryptographic hash varies enormously with its input and exercises no discretion, because at each input the commitments fix the output exactly. It is not an observer's inability to predict a fixed policy, which is a fact about the observer and not about the arrangement. And it is not randomness: a deterministic arrangement can carry substantial discretion across unfamiliar cases, a randomised one can be tightly committed, and a zero-variance arrangement can be consistently wrong. | +| `term:seam` | seam | The **seam** — the coordination boundary decomposition creates between the parts: splitting a task or an actor manufactures new governing decisions, the ones about how the parts coordinate, that did not exist when the thing was whole. A composite carries the demand of its parts, **plus** the seam demand `S` created *between* them. | +| `term:store` | store | **{rule, check, actor, nothing}.** There is no fifth source. | +| `term:swarm-gate` | swarm-gate | **A swarm is an actor only if it genuinely determines choices against ground.** The admission tests (`00` §4) still gate, and they must. | +| `term:training` | training | **Closure decides whether training is *available*. Cost decides the *ratio* when it is.** Training requires a **reliable error signal** — you must be able to tell, and tell soon, whether the output was right. | +| `term:verdict` | verdict | **Definition (determination demand).** For a task whose acceptance predicate **closes** for the arrangement (`term:closure`; *decidable* is the formal special case, not the requirement), the predicate evaluates outcomes, and the **task class** supplies one correct output per input point. The **verdict** is that induced assignment — the correct output over each point of the input space. Let `P` be the distribution over inputs (the *ground distribution*). The **determination demand** of the task is the Shannon entropy of the verdict: **D = H(verdict)**, measured in **bits**. Where the task class supplies no such assignment, the predicate still evaluates outcomes and there is no verdict to have entropy about — which is the boundary `09` §7 draws. | + +### The hypothesis set, as the graph holds it + +The hypothesis set is broken out because its discipline is the easiest thing in the paper for a +reader to mistake. Every row is `projected`, every row declares a falsifier, **every evidence field +is empty**, and every row is owned by a study that has not been run. The columns below are the +graph's own fields, not the paper's summary of them. + +| ID | Status | Evidence | Owner | Falsifier declared | +|---|---|---|---|---| +| `DDD-frame-07` | projected | `[]` (empty) | paper-4 | yes | +| `DDD-hyp-01` | projected | `[]` (empty) | paper-4 | yes | +| `DDD-hyp-02` | projected | `[]` (empty) | paper-4 | yes | +| `DDD-hyp-03` | projected | `[]` (empty) | paper-4 | yes | +| `DDD-hyp-04` | projected | `[]` (empty) | paper-4 | yes | +| `DDD-hyp-05` | projected | `[]` (empty) | paper-4 | yes | + +*Generated from the graph at `v5.12.0`. 48 claims, 3 decisions, 24 terms.* diff --git a/papers/paper-a/paper-a.md b/papers/paper-a/paper-a.md index 69ce3e7..c938532 100644 --- a/papers/paper-a/paper-a.md +++ b/papers/paper-a/paper-a.md @@ -4,13 +4,13 @@ *Emil — Context&. Statement paper.* -*This paper is a projection of `actor-indexed-determination` at `v5.9.0`. Every bracketed claim, -decision and term identifier resolves against that one ref, and Appendix A reproduces each of them -from it, so the paper can be checked without the repository. No identifier here resolves against the -downstream repository, so no downstream ref is pinned. The companion measure note — -`papers/measure-note/measure-note.md` in `decision-driven-design` — is cited by path at commit -`aa7e135`, which carries its discharge section; the note is absent from `v0.4.0` and from every -earlier downstream tag, so it resolves at that commit rather than at a tag.* +*This paper is a projection of the `actor-indexed-determination` claim graph at `v5.12.0`. Every +bracketed identifier resolves against that one ref, and the supplement reproduces each cited node +word-for-word from it, so the paper can be checked without the repository.* + +*The supplement — `paper-a-supplement.md`, beside this file — carries the reproduced nodes, the +projection's mechanics and the reproduction instructions. It is not required to follow the +argument.* --- @@ -36,8 +36,10 @@ never unmet, only ungoverned**; and discharge is act-indexed, so governance neve demand is supplied, only by what. On the region where the acceptance predicate closes, demand is identified with the entropy of the verdict, and on that identification conservation is the chain rule of entropy. The identification is a modelling claim and it is unproven; the chain rule is -Shannon's and is certain and empty on its own. Off the closing region the measure does not exist -and the governance question still does. +Shannon's and is certain and empty on its own. **The construction has three conditions, not one** — +a verdict function must exist, the evaluation must be available to the arrangement, and the +deployment distribution must be estimable — and closure governs only the second. Where the +construction does not apply, the governance question still does. **What is claimed is a synthesis, not a discovery.** That consequential work should be analysed at the level of a sociotechnical arrangement rather than an isolated actor is not this framework's @@ -52,19 +54,27 @@ what is merely proposed. No empirical result is claimed, and none is reported. ## Note on claim status This paper is a **projection**. Every load-bearing statement below either cites a node in the -framework's claim graph or is marked as the paper's own analysis. The graph's status vocabulary is -used throughout and is not this paper's to redefine: - -- **settled** / **established** — argued and unchallenged in the graph. -- **reported** — exercised by a reproducing computation. -- **projected** — proposed with a declared falsifier, not yet met. -- **draft** — filed, not yet ratified. -- **retired** — superseded, retained with the correction that killed it. - -The paper's primary claim is `projected` [DDD-frame-01]. So is most of what follows. Where a -statement is the paper's own analysis rather than the graph's, it is marked **(analysis)** and -carries no claim status at all, because it is not a claim. Appendix A reproduces every cited node -word-for-word from the graph, with its status, so the paper can be checked without it. +framework's claim graph or is marked as the paper's own analysis. + +**What the status labels do and do not mean.** They record how far a proposition has been argued +**inside this framework**, and nothing else. `settled` and `established` mean argued and +unchallenged here — **not empirically confirmed, and not externally validated**. `reported` means a +computation runs and reproduces, not that the world was consulted. `projected` means proposed with a +declared falsifier that has not been met. `draft` means filed and not yet ratified. `retired` means +superseded and kept with the correction that killed it — and this paper carries three of those, two +of them killed by the review it answers. + +The paper's primary claim is `projected` [DDD-frame-01, projected]. So is most of what follows. +Where a statement is the paper's own analysis rather than the graph's, it is marked **(analysis)** +and carries no status at all, because it is not a claim — which does not make it uncontestable, only +unattributed to the graph. + +An external review found this vocabulary being read as epistemic warrant when it encodes internal +maturity. **That reading is wrong and the paper's earlier phrasing invited it.** The supplement +reproduces every cited node with its status and its kind, so a reader can check the pairing rather +than take this paragraph's word for it: every `established` claim in this graph is `formal`, so what +is settled here is arithmetic, and the modelling claims that give the arithmetic its meaning are all +`projected`. --- @@ -115,8 +125,10 @@ trusted, who answers, and which residual risks stay accepted — or escape. The paper claims a framework and a vocabulary, together with a set of graded predictions that are not yet tested. It does not claim an empirical result. It does not claim a boundary between human and computational work. It does not claim that determination demand is a measured invariant in -general — only on the region where the acceptance predicate closes, where the measure exists and -the conservation statement is arithmetic (§4.4). +general — only where the verdict-entropy construction is available to the arrangement, which +requires the acceptance predicate to close **and** the task to supply a verdict function **and** the +deployment distribution to be estimable; on that region, and given the identification, the +conservation statement is arithmetic (§4.4). **And it does not claim the arrangement.** An earlier version of this paper was titled *The Missing Parameter*, which asserts absence from prior work. The absence does not hold. Analysing @@ -263,11 +275,9 @@ material that would settle it is unfiled, and this paper does not anticipate it. provenance is named here and its mechanism is stated as pending that filing, cited as an open question rather than smuggled in early. -*This five-way partition is not canon. The framework has ruled it ineligible for minting rather than -merely deferred — a mint would fix a partition that the open trust question may restructure — -so it is carried here as this paper's analysis, marked non-canon at the point of use, with the -institutional slot's mechanism flagged as pending on that same question [DDD-dec-26]. This paper -flags the gap and files nothing.* +*This five-way partition is **not canon**. It is this paper's analysis, marked non-canon at the point +of use, and §8.1 shows a case it does not adjudicate. The framework's reason for not minting it is +recorded in the supplement.* ### 2.5 The arrangement, not the isolated actor @@ -293,12 +303,11 @@ How does an arrangement fix behaviour in advance? The framework's answer is thre levels compose. > **Behavioural commitments attach at three levels — outcome, policy, principal — which compose and -> are not actor species; residual discretion is the outcome-relevant variation that remains after -> the arrangement's declared commitments are applied.** [DDD-frame-02, projected] +> are not actor species; residual discretion is the outcome-relevant variation those commitments +> leave open at the act, held at fixed ground.** [DDD-frame-02, projected] -Two terms name the halves of that claim. Both were minted only when a downstream projection needed -them as citable nodes, and **both carry `draft` status** — they are filed and not yet ratified, and -this paper shows the status rather than quietly promoting it [DDD-dec-26]. +Two terms name the halves of that claim, and **both carry `draft` status** — filed, not yet +ratified. The paper shows that rather than quietly promoting it. ### 3.1 The three levels @@ -328,19 +337,39 @@ the error the level vocabulary exists to prevent. ### 3.2 Residual discretion -> **Residual discretion** is the outcome-relevant variation remaining at the act after the -> arrangement's declared commitments are applied. It is not randomness: a deterministic arrangement -> can carry substantial discretion across unfamiliar cases, a randomised one can be tightly +> **Residual discretion** is the outcome-relevant variation the arrangement's commitments +> leave open at the act, **held at fixed ground**: the alternatives still admissible once +> the standing configuration and the ground at the act are both given. It is not variation +> across ground — a cryptographic hash varies enormously with its input and exercises no +> discretion, because at each input the commitments fix the output exactly. It is not an +> observer's inability to predict a fixed policy, which is a fact about the observer and +> not about the arrangement. And it is not randomness: a deterministic arrangement can +> carry substantial discretion across unfamiliar cases, a randomised one can be tightly > committed, and a zero-variance arrangement can be consistently wrong. > [term:residual-discretion, **draft**] -Those three clauses each block a different mistake. Determinism is not commitment: a program that -always does the same wrong thing on an unfamiliar input has enormous outcome-relevant variation -across the input space, all of it unresolved by any declared commitment. Randomisation is not -discretion: a sampler drawing from a tightly bounded distribution the arrangement declared and -accepted has been committed at policy level. And low variance is not correctness: an arrangement -can be consistent and consistently unacceptable, which is why variance is never the measure of -commitment. +**The held-at-fixed-ground clause is a repair, and the earlier version of this paper carried the +defect it repairs.** An external review found the term conflating four phenomena: outcome variation +across ground, epistemic uncertainty about a fixed policy, stochasticity, and genuine unresolved or +delegated selection. Its decisive example is the one canon now carries — **a cryptographic hash +varies enormously across inputs and exercises no discretion whatever**, because at each input the +commitments fix the output exactly. Under the earlier wording, "the variation remaining after the +declared commitments are applied", the hash came out carrying enormous discretion, which is plainly +wrong. Quantifying at *fixed ground* is what excludes it. + +Four clauses now block four different mistakes, and only the last was there before. **Variation +across ground is not discretion** — the hash. **Unpredictability is not discretion**: an observer +who cannot anticipate a deterministic policy has learned something about the observer, and the +arrangement's commitments are what they were. **Randomisation is not discretion**: a sampler drawing +from a bounded distribution the arrangement declared and accepted has been committed at policy +level. And **low variance is not correctness**: an arrangement can be consistent and consistently +unacceptable, which is why variance is never the measure of commitment. + +**(analysis)** What remains after the separation is narrower and more useful: discretion is the set +of alternatives still admissible when both the standing configuration and this act's ground are +given. That is the quantity §4.1's fixing test reads — an alternative with no residual discretion at +fixed ground is *predetermined* [DDD-frame-17, projected] — so the two repairs the review forced +turn out to be one repair, made in two nodes. ### 3.3 Boundary cases @@ -378,21 +407,58 @@ That separation is the section's spine. Section 4.1 takes the source side, §4.2 §4.3 what happens when neither is adequate, and §4.4 what is conserved across every rearrangement of the two. -### 4.1 Source of resolution: the four discharge modes +### 4.1 Source of resolution: the three discharge values A completed act's determination demand is not sometimes supplied and sometimes not. It is always supplied, and the only question is by what. -> **At every completed act in a task's scope, the act's determination demand is discharged — by a -> filed decision, an actor's judgment, an arrangement default, or an uncontrolled draw; escape is a -> supply mode of discharge, not an absence of supply — demand is never unmet, only ungoverned.** -> [DDD-frame-15, projected] - -The four modes classify **the producer of the resolution**, not its governance status. A *filed -decision* is variation authored in advance and delivered at the act. *Judgment* is variation -produced at the act by an actor reading ground. A *default* is variation carried by the -arrangement's standing configuration without a fresh resolution — and a default may be declared and -governed, or incidental and not. A *draw* is variation living in nothing the arrangement controls. +> **At every completed act in a task's scope, each outcome-relevant alternative is discharged in +> exactly one of three ways: predetermined — the arrangement's standing configuration together with +> the ground at the act determines the resolution; exercised — it does not, and something within +> the arrangement's control determines it at the act; or drawn — it does not, and what determines +> it lies outside the arrangement's control. Demand is never unmet, only ungoverned.** +> [DDD-frame-17, projected] + +**This claim is three days younger than the objection that produced it, and the paper says so.** +An earlier version of this section carried a four-mode enumeration — filed decision, judgment, +arrangement default, uncontrolled draw — which an external review showed does not partition: a +declared default is both a filed decision and an arrangement default, a thermostat satisfies the +judgment gloss under a rule canon times as encoded, and trained inference is policy commitment, +judgment and standing encoded supply at once. The enumeration was retired in canon rather than +defended [DDD-frame-15, retired; superseded by DDD-frame-17]. The paper carries the successor. + +The three values turn on **two dichotomies and nothing else**. First: do the arrangement's standing +configuration and the ground at the act, together, fix the resolution? If yes it is +**predetermined**. If no, second: is what fixes it inside the arrangement's control? Inside is +**exercised**; outside is **drawn**. + +Three things follow, and each repairs one of the overlaps. + +**The unit is the alternative, not the act** [DDD-frame-17, region]. One act may carry several +outcome-relevant alternatives that discharge differently — a timeout carries a when-to-stop +alternative and a what-to-return alternative, and they need not land the same way. The retired +claim quantified over acts in its statement and over alternatives in its own test field, which are +different claims; fixing the unit dissolves several reported overlaps before any of the rest is +needed. + +**Governance status is not an axis of this claim** [DDD-frame-17, region]. Whether a resolution is +declared, governed, filed or escaped is [term:store]'s question and is answered there. The three +values turn on determination and control alone, never on declaredness — which is what makes the +seam below an exclusion the claim's own shape enforces rather than a paragraph a reader may skip. +The declared-default overlap was exactly a declaredness test smuggled into a locus partition. + +**No value turns on actorhood** [DDD-frame-17, region]. The fixing test asks whether standing +configuration and ground jointly determine the resolution, which is answerable without deciding +whether the producer is an actor. The retired claim's *judgment* mode did turn on actorhood, and +thereby inherited the open circularity in the actor admission test (§2.1, and the review is right +that it is open). This claim does not inherit it. + +**(analysis)** The value is a property of the arrangement at the act, not of the carrier in the +abstract, and the same weights can take different values: a carrier decoding greedily is +predetermined, and the same model sampled at temperature is exercised — or drawn, where the +arrangement does not control the sampling source. An arrangement that changes its decode changes +the value without changing anything it bought. Coding practice must therefore read the deployment +and never the artefact, which is a real burden on §9's Study 0 and is not a small one. Discharge is also indexed to the act, on both sides of the supply split: @@ -414,21 +480,33 @@ rather than left to be inferred. | | The store partition | The discharge partition | |---|---|---| | Asks | what **governance supplied** | what **the world produced** | -| Values | `{rule, check, actor, nothing}` [term:store] | filed decision · judgment · default · draw [DDD-frame-15] | +| Values | `{rule, check, actor, nothing}` [term:store] | predetermined · exercised · drawn [DDD-frame-17] | | Escape is | **nothing** — "there is no fifth source" | **a supply mode** — the uncontrolled draw | -Both are correct, and neither reduces to the other [DDD-frame-15, region]. In the store partition +Both are correct, and neither reduces to the other [DDD-frame-17, region]. In the store partition escape is nothing, because the question is what governance supplied and the answer is that nothing -did. In the discharge partition the same act's demand is met by an uncontrolled draw, because the -question is what the world produced, and **the world never produces nothing**. A reader who reads -one partition's "nothing" into the other's four modes will conclude the framework contradicts -itself; it does not, and the two questions are simply distinct. - -*This paper's §4.1 replaces the source enumeration its own foundation document used —* prior -commitment, runtime actor, environmental-or-default, failure-or-non-resolution. That enumeration -mixed declared defaults with uncontrolled dynamics in its third source, and its fourth was not a -discharge mode at all, since even a failed arrangement's act lands an outcome in the world register -[DDD-frame-15, notes]. Canon governs, and the paper carries canon's form. +did. In the discharge partition the same alternative is *drawn*, because the question is what the +world produced, and **the world never produces nothing**. A reader who reads one partition's +"nothing" into the other's three values will conclude the framework contradicts itself; it does +not, and the two questions are simply distinct. + +**This guard survived the retirement, and that is worth recording.** The seam was ratified with the +four modes and killed with none of them: it separates governance-supply from discharge, which is +orthogonal to the overlaps that failed, and it is carried into the successor's region verbatim +[DDD-frame-17, region]. It is also the one part of that section an external reader met and still +found confusing, which is a finding about exposition rather than about the guard. + +*Two enumerations have now been retired at this seam, and both for the same reason.* The +foundation document's — prior commitment, runtime actor, environmental-or-default, +failure-or-non-resolution — mixed declared defaults with uncontrolled dynamics in its third source, +and its fourth was not a discharge mode at all, since even a failed arrangement's act lands an +outcome in the world register [DDD-frame-15, notes]. Its replacement, the four modes, mixed causal +producer with provenance, timing, governance status and control. **A list of sources drawn from +several dimensions at once will not partition**, and the successor's answer is to stop enumerating +sources and dichotomise instead: determined or not, and if not, controlled or not. Whether *that* +partitions is `projected` and has a falsifier — an alternative for which there is no fact of the +matter whether standing configuration and ground determine the resolution [DDD-frame-17, +falsifier]. Canon governs, and the paper carries canon's form. ### 4.2 Assurance mechanism @@ -472,7 +550,7 @@ side. It follows that **escape is tested on the arrangement, not on the outcome**. Ask whether any adequate source-and-assurance combination governed the act. If none did, the act escaped whether or not what landed happened to be acceptable, because what landed was drawn from a distribution the -arrangement did not control [DDD-frame-15, derivation]. +arrangement did not control [DDD-frame-17, projected]. #### Filing is not delivering @@ -548,11 +626,33 @@ for two decompositions of one task: totals of 25.493 bits either way, split 20.5 decomposition and 11.020 / 14.474 by the other (measure note §4; `core/09` §3). Demand is conserved across the two, not escaped by either. -**The bound, stated and not implied.** The measure exists **if and only if** the acceptance -predicate operationally closes, and `H(V)` is undefined exactly where the framework locates a -non-zero floor [DDD-measure-06, **established**]. Off the closing region there is no verdict -function to have entropy about, and conservation is what it was before the measure existed: an -accounting discipline, a principle rather than a measured invariant. +**The bound, restated after a correction, and it was the paper's largest overstatement.** An earlier +version of this section said the measure exists **if and only if** the acceptance predicate +operationally closes, citing a claim then carried at `established`. **That biconditional is false in +both directions, the node asserting it is retired, and the correction came from outside** +[DDD-measure-06, retired]. + +Three conditions were collapsed under one, and they fail differently [DDD-measure-16, established; +DDD-measure-17, projected]. + +| Condition | What it asks | What closure does about it | +|---|---|---| +| **Existence** | does the task class supply a unique correct output per input point? | **nothing.** A relation task can have a fully decidable acceptance predicate and admit several acceptable outputs, and no output-valued `V` exists until the task supplies a tie-breaker, a canonical form, or a declared selection distribution | +| **Availability** | can *this arrangement* run the evaluation within its declared bounds? | **this is what closure is.** It is a property of the arrangement, not of the task, and the same nominal task can be inside the construction's domain for one arrangement and outside it for another | +| **Estimability** | is the deployment distribution known well enough to compute the quantity? | **nothing.** Where it is unknown, unstable, non-stationary or partly observable, the demand is well defined and cannot be computed | + +So closure governs **availability** and neither of the other two [DDD-measure-16, **established**]. +Off the closing region this construction does not apply — which is not the same as measurement +failing there, and the surviving claim is careful to say so: where evaluators disagree there are +distributions over their judgments, and some of those are informative [DDD-measure-16, region]. + +**(analysis)** The provenance of this repair is worth stating, because the paper's credibility rests +more on it than on the repair. The companion measure note's own revision had already conceded both +failure directions in its scope condition, and canon was never updated to match — so this projection +was *stronger than the note that shares its graph*, and stayed so through a merge. An external +review found it. The node then retired at `v5.10`, and this section is the projection catching up. +The instrument that should have caught it did not: `DDD-measure-06` was not pinned, so no warning +fired when its status moved (§ the reproduction note). **Governance outruns measurement.** The governance question — is every decision governing the act in a declared store, none escaped? — is well-formed on the total domain, including open predicates @@ -626,15 +726,21 @@ assembled in one place. **Three rungs, weakest first**, and all three are operat | **Verification-closed** | Adequacy can be *evaluated* within declared resource, latency and confidence bounds. | [term:closure] | | **Constructively closed** | The verdict is *computed by rule* from ground available at the act. | measure note §8 | -**Open.** No procedure decides acceptability, so there is no verdict function to have entropy about -and the measure does not exist [DDD-measure-06, **established**]. Governance is untouched by this: +**Open.** No procedure decides acceptability over accessible ground within declared bounds, so the +verdict-entropy construction is not available to this arrangement [DDD-measure-16, established]. +Two things it does *not* mean, and the retired predecessor blurred both: it does not mean no verdict +function exists, and it does not mean nothing about the situation is measurable +[DDD-measure-16, region]. Governance is untouched by either: the question of whether every decision governing the act sits in a declared store, none escaped, is well-formed on the total domain [DDD-frame-11]. The measure's silence here is a fact about the measure. **Verification-closed.** The rung canon means by *closure*: the relevant ground is observable and adequacy can be evaluated within declared bounds [term:closure]. Two things become available at -once. The measure exists, so specification demand is `H(V)` [DDD-measure-01]. And producer identity +once. The verdict-entropy construction becomes available to the arrangement — **available, which is +not the same as defined or computable**, and the other two conditions still have to hold — so where +they do, specification demand is `H(V)` [DDD-measure-01, projected; DDD-measure-16, established]. +And producer identity stops being epistemically necessary — for the checked property, and nothing more [DDD-frame-05]. **Constructively closed.** A predicate can be more than checkable. Call it **constructively closed** @@ -683,13 +789,11 @@ no generation cost to bound, and predicates whose verdicts are merely checkable scoped survivor of the retirement — that producer identity is not necessary *for the checked property and nothing more* — holds either way [DDD-frame-05]. -*Canon's closure vocabulary does not carry the constructive/verification distinction. `term:closure` -is stated in evaluative terms alone, no claim node names the stronger rung, and the word* -constructive *occurs nowhere in the principle repository's core documents. The rung above is stated -register-native, as this paper's analysis and not as canon; a dedicated node is pending filing on -the framework's open-questions wave, and until it lands the citation basis is closure, the -separation of closure from generation cost, and the measure's silence on search, as cited above. -This paper flags the gap and files nothing.* +*Canon's closure vocabulary does not carry the constructive/verification distinction: `term:closure` +is stated in evaluative terms alone and no claim node names the stronger rung. **The constructively +closed rung is therefore this paper's own, not the graph's**, and it is stated register-native. Its +basis is closure, the separation of closure from generation cost, and the measure's silence on +search, all cited above. The filing status is recorded in the supplement.* ### 5.3 Closure is not generation cost @@ -704,7 +808,7 @@ space. None of these contradicts closure. What they refute is the inference *fro cheap adequacy, and that inference is the one the framework retired a claim over (§10). The measure register says the same thing from its own side: **the measure prices the verdict, not -the search.** `H(verdict)` is a property of the verdict function and the ground distribution, and +the search.** `H(verdict)` is a property of the verdict function and the deployment distribution, and two tasks with identical verdict entropy can differ without bound in what it costs to compute a correct answer [DDD-measure-11, reported]. @@ -789,16 +893,21 @@ Polanyi's and Collins's, and this paper cites them for the phenomenon only, taki either (Polanyi 1966; Collins 2010). What is the framework's, and not theirs, is the residue's *location*: in the checkability of the acceptance predicate, arrangement-indexed. -The measure's boundary coincides with the floor's location: **the measure exists if and only if the -acceptance predicate operationally closes, and `H(V)` is undefined exactly where the framework -locates non-zero floor** [DDD-measure-06, **established**]. +The construction's domain and the floor's location coincide, and canon now carries that coincidence +as its own claim rather than folding it into the boundary: -**That coincidence is worth noticing and it is not evidence.** The two arguments share a premise — -the closure of the acceptance predicate — so their agreement about where the line falls is close to -definitional on the measure's side. What is not definitional is that the line was drawn twice, from -different materials, with neither drawing fitted to the other. That makes the boundary principled -rather than arbitrary. It does not make the identification true, and this paper does not argue that -it does (measure note §8.2). +> **The verdict-entropy construction's domain and the region where the framework's floor result +> locates non-zero floor coincide, and the coincidence is principled rather than evidential: the two +> arguments share the closure of the acceptance predicate as a premise, so their agreement about +> where the line falls is close to definitional on the measure's side and is not evidence that the +> identification is correct.** [DDD-measure-17, projected] + +**That the claim is `projected` rather than `established` is the point of it.** Its predecessor +carried this content at `established`, and the content includes a judgment about what the agreement +is *worth* — which is an interpretive claim about two arguments, not arithmetic. What is not +definitional is that the line was drawn twice, from different materials, with neither drawing fitted +to the other, and that makes the boundary principled rather than arbitrary. It does not make the +identification true, and this paper does not argue that it does (measure note §8.2). ### 6.2 The capacity mechanism, and its scope @@ -982,7 +1091,7 @@ nothing about the size of that residual [term:residual-discretion]. The last two rows are the section's point. Both acts complete and both land outcomes in the world; one is governed by an arrangement default nobody declared, and one is not governed at all. Under -[DDD-frame-15] both are discharged — demand is never unmet, only ungoverned — and under [term:store] +[DDD-frame-17] both are discharged — demand is never unmet, only ungoverned — and under [term:store] the last row's source is *nothing*, which is an escape. The second row carries the qualification: if the model's context does not deliver that filed decision at the act, no source supplied it, and it escapes too [DDD-delivery-02]. @@ -995,7 +1104,7 @@ substantially weaker statement than conformance to intent. Canonical-form correc **constructively closed** for the countries whose rules are encoded — the correct output is computed by rule from institutional ground, and there is no search to price (§5.2). Behaviour on an uncovered country is **open**: no procedure over accessible ground decides acceptability, and there -is nothing for `H(V)` to be about [DDD-measure-06]. +the verdict-entropy construction is not available to this arrangement [DDD-measure-16, established]. Closure varies *within a single act*, per capability, which is precisely why required actor class is computed per capability rather than per act [DDD-cost-12]. @@ -1219,8 +1328,17 @@ arithmetically valid split with no engineering meaning, and the condition is wha *The escape mechanism's scope.* Overflow ∩ open is sufficient for escape and never necessary for it (§6.2) [DDD-dec-15]. -*The measure's silence.* Off the closing region the measure does not exist, and its absence licenses -no claim about governance, which is well-formed there [DDD-frame-11]. +*The measure's three conditions.* Closure governs the construction's **availability** to an +arrangement and neither the verdict function's existence nor the deployment distribution's +estimability [DDD-measure-16, established]. Where the construction does not apply its silence +licenses no claim about governance, which is well-formed there [DDD-frame-11], and none about +measurement in general [DDD-measure-16, region]. + +*A retirement this paper carried into print before it was caught.* The predecessor of that claim +asserted a biconditional in both directions at `established`, this paper projected it faithfully, +and the companion measure note had already conceded both failure directions without canon being +updated to match (§4.4). The projection was stronger than the note sharing its graph. **An external +review found it; no instrument in this repository did.** ### 10.3 Boundary cases @@ -1498,12 +1616,13 @@ adequate commitment, check or accountable authorisation governs [term:escape]. The framework's compact statement of what makes that possible is the conjunction of two claims: -> **… demand is never unmet, only ungoverned.** [DDD-frame-15 — closing clause] +> **… demand is never unmet, only ungoverned.** [DDD-frame-17 — closing clause] -That sentence is canon's own, and the compact statement it belongs to is the **conjunction** of two -claims rather than either alone: every completed act's demand is discharged by one of four modes -[DDD-frame-15], and discharge is act-indexed, so governance never chooses whether demand is -supplied, only by what [DDD-frame-16]. +That sentence is canon's own, and it survived the retirement of the claim it was first filed in +(§4.1): the four-mode enumeration died and the compact form did not. The statement it belongs to is +the **conjunction** of two claims rather than either alone: every outcome-relevant alternative at a +completed act is discharged in one of three ways [DDD-frame-17], and discharge is act-indexed, so +governance never chooses whether demand is supplied, only by what [DDD-frame-16]. And its most load-bearing proposal is relational — a proposal, with a falsifier, and without evidence: @@ -1513,9 +1632,10 @@ evidence: > [DDD-floor-02 — opening clause; the claim continues, quoted in full at §6] **What is established, and what is projected.** Established in the graph: that closure is distinct -from generation cost; that conservation on the closing region is the chain rule; that the measure -exists exactly where the acceptance predicate closes; that the parts of a decomposition become -trivial only when the decomposition already encodes the verdict. Projected, with falsifiers and +from generation cost; that conservation on the region where the construction applies is the chain +rule; that the construction's availability to an arrangement is what closure governs, and not the +verdict function's existence or the distribution's estimability; that the parts of a decomposition +become trivial only when the decomposition already encodes the verdict. Projected, with falsifiers and without evidence: the index itself, the commitment levels, the discharge chain, the relational floor, the accountability relation, and every one of the five hypotheses. **Every claim in the first list is `formal`** — arithmetic, or a derivation from it — so what is settled here is arithmetic, and @@ -1542,22 +1662,6 @@ from. --- -## Reproduction - -**This paper mints no figures and no assets.** Every number it states was produced by an existing -script in the principle repository and is cited to the projection that works it: the -date-validation totals of §4.4 and §8.6 come from `core/assets/measure-toy.py`, worked in the -companion measure note's §4 and in `core/09` §3. That script was re-run fresh while this paper was -drafted and reproduces the stated values. - -Every block quotation attributed to a graph node was verified verbatim against the graph at the -pinned ref by `check-quotations.py`, alongside this file. Appendix A was generated from the graph by -`gen-appendix.py` and re-read against it by `check-appendix.py`. All three scripts live beside this -manuscript and take the ref as an argument, so the checks are reproducible against any ref rather -than against the one that happened to be current. - ---- - ## References Locators are marked **verified** where the locator was checked against an authoritative registry — @@ -1661,133 +1765,3 @@ Engineering*. CRC Press, Boca Raton. ISBN 0-8493-3933-2, doi:10.1201/97814200056 volume the external review's own locator resolves to**, while the review's prose concerns the 2005 *Foundations* volume above. *(Locator verified against Crossref and Open Library; the volume itself was not obtained.)* - ---- - -## Appendix A. Cited claims, decisions and terms - -The paper cites nodes in the framework's claim graph. Statements below are reproduced -word-for-word from the graph at the ref pinned in the front matter, so the paper can be checked -without it. **Kind** and **status** are the graph's own fields, and they answer different questions. -*Kind* is what sort of claim it is: *formal* is arithmetic or a derivation, *empirical* rests on -observation, *conceptual* fixes or uses the framework's vocabulary, *normative* says what ought to -be done. *Status* is how far it has been argued: *settled* and *established* are argued and -unchallenged **within this framework**, *reported* is exercised by a reproducing computation, -*projected* is proposed with a declared falsifier and not yet met, *draft* is filed and not yet -ratified, *retired* is superseded and kept with the correction that killed it. - -**Neither field claims external validation, and the two must be read together.** *Established* -means internally argued and unchallenged, not empirically confirmed; *reported* means a computation -runs and reproduces, not that the world was consulted. The pairing is what carries the information: -every *established* claim in this graph is *formal*, so what is settled here is arithmetic, and the -modelling claims that give the arithmetic its meaning are *projected*. The canonical statement of -what each value means, and what it does not, is `spec/claim-format.md` §5 at the pinned ref; this -paragraph projects it and does not replace it. - -**This appendix is generated from the graph and never hand-edited** (`gen-appendix.py`), then -re-read against the graph by an independent script (`check-appendix.py`). - -### Claims - -| ID | Kind | Status | Statement | -|---|---|---|---| -| `DDD-agent-01` | empirical | projected | Long-running agent drift is escaped decisions caused by basis loss: context decay, compaction, and distractors remove claim nodes from the agent's ground, so later actions are decisions with no basedOn edge to the declared claims and revert to model priors. Grounding the agent in a persistent external claim graph — basis as query, not context residue, with per-decision claim citation — reduces drift relative to context-carried instruction. | -| `DDD-cost-08` | conceptual | projected | Actor selection for an act is two-gated: capacity gates always — the actor must carry the act's residual at the declared tolerance or the excess escapes — and assurance gates exactly where the acceptance predicate does not close, where assurance must attach to the actor because no check can carry it. | -| `DDD-cost-09` | conceptual | projected | Assurance-by-actor binds assurance to a scarce carrier, supplied occasioned — at the act, at the carrier's class price. Assurance-by-check moves it into a mechanism, supplied standing — independent of the act. Closing a predicate converts a property's assurance supply from occasioned to standing. | -| `DDD-cost-11` | conceptual | projected | On an open predicate, assurance and actor class are positively coupled; closing the predicate flips the sign — the assurance gate lifts and the capacity gate softens — so the actor class the act requires falls, leaving the actor carrying generation only. | -| `DDD-cost-12` | conceptual | projected | Required actor class for an act is the maximum, over the act's capabilities, of the class needed where assurance is not mechanically discharged — per capability, not per act. | -| `DDD-cost-13` | conceptual | projected | Where assurance attaches to the actor per capability, an answer-keyed qualification instrument — an examination, an eval — evidences demonstrated class on predicates that close. Its verdict cannot evidence open-predicate carriage: the instrument's own predicate closes while the target predicate does not, so delegation to the open predicate substitutes the actor's identity for exactly the check the instrument cannot be. | -| `DDD-cost-20` | conceptual | projected | Encoding around a carrier and encoding within it differ in locus: around-encoding — context, retrieval, scaffolds — is standing supply outside the carrier, delivered through the channel at each act; within-encoding — training — converts judgment allocation to encoded allocation inside the carrier. Training buys allocation, not capacity: it does not enlarge the judgment store. | -| `DDD-cost-25` | conceptual | projected | Assurance mechanisms occupy temporal positions relative to the act — pre-act (selection, training, encoding, static checks), at-act (monitoring), post-act (review, audit, consequence) — each with a latency; a mechanism whose latency exceeds its position's budget — the episode for at-act mechanisms, the consequence horizon for post-act ones — cannot hold its position, so rising tempo, which compresses both budgets, forces assurance pre-act, into standing supply or the carrier. | -| `DDD-delivery-01` | conceptual | projected | Filing is not encoding: a decision sits in I(V;X) only to the extent the arrangement delivers it at the act, so store allocation cannot be read off artefacts, and the paid-once-inherited-by-every-run property belongs to mechanical delivery specifically, not to standing supply generally. | -| `DDD-delivery-02` | conceptual | projected | Governance filed but not delivered is escape: no source supplied the governing decision at the act, so it was determined by nobody (term:escape, supply-general), and delivery failure is thereby a generator of escape — sufficient, never necessary — whose distinguishing feature is that the ledger shows coverage: escape that presents as governance. | -| `DDD-delivery-03` | conceptual | projected | An unretrieved decision and an unretrieved check over the same act are correlated failures — same actor, same budget, same position — so judgement-mediated delivery on both the source and assurance sides silently removes the independence a gate depends on: the failures compound rather than stack. | -| `DDD-floor-01` | formal | reported | H(V\|X) bundles judged and escaped demand; cleaving them requires an actor-capacity model, and residual demand an actor has taken up escapes where it exceeds effective capacity min(C_hold, C_resolve) AND the shed decisions carry no verifier — overflow ∩ open is the mechanism of capacity-generated escape, sufficient for escape and not necessary for it, with overflow alone producing retries, not escape. | -| `DDD-floor-02` | conceptual | projected | The judgment floor is relational: irreducibility is a property of the indexed relation ⟨task, ground, acceptance relation, tolerance, arrangement, assurance⟩, not of the task alone — the portion of determination an arrangement cannot discharge through its prior commitments or adequate direct verification at the declared assurance level moves when any coordinate of the relation moves. | -| `DDD-frame-01` | conceptual | projected | Unresolved determination is indexed by the tuple ⟨task, ground, acceptance relation, tolerance, arrangement, assurance⟩, not by the task alone. | -| `DDD-frame-02` | conceptual | projected | Behavioural commitments attach at three levels — outcome, policy, principal — which compose and are not actor species; residual discretion is the outcome-relevant variation that remains after the arrangement's declared commitments are applied. | -| `DDD-frame-03` | conceptual | projected | The source of a resolution and the mechanism assuring it are separate dimensions; the four-store model held the arrangement largely fixed while the allocation was analysed, and did not draw the distinction. | -| `DDD-frame-04` | empirical | projected | Escaped decisions — consequential resolutions with no adequate source-and-assurance combination — predict ungoverned failure modes and design-review findings. | -| `DDD-frame-05` | formal | projected | Under a sound terminating operational checker with complete declared ground, producer identity is not epistemically necessary for the checked property — and nothing more: not cheap generation, not normative completeness, not accountability. | -| `DDD-frame-06` | formal | established | Closure is distinct from generation cost: verification being cheap implies nothing about the density or accessibility of the acceptance region. | -| `DDD-frame-07` | empirical | projected | Operational evaluability, feedback density, and ground accessibility predict the comparative advantage of computationally assisted arrangements over unaided situated judgment, after controlling for difficulty and resources (H1–H5, filed as DDD-hyp-01 through DDD-hyp-05). | -| `DDD-frame-08` | normative | projected | Accountability is a relation (attribution, persistent principal, authority linkage, stake, sanction path), not an intrinsic capacity; an arrangement naming an executor but no principal is incomplete. | -| `DDD-frame-09` | formal | retired | RETIRED — "closed predicates make intelligence unnecessary." Does not follow from producer-independence under verification; generation may still require whatever capability the search demands. | -| `DDD-frame-10` | formal | projected | Determination demand is conserved as a scalar across arbitrary re-arrangements, actor-generally and including open predicates. | -| `DDD-frame-11` | conceptual | projected | The governance question — is every decision governing the act in a declared store, none escaped? — is well-formed on the total domain, including open predicates where the measure does not exist; the cost question — how much is in each store? — exists only where the predicate closes. The framework's governed domain is strictly wider than its measured domain. | -| `DDD-frame-13` | conceptual | projected | Determination's object is the determinable — an outcome-relevant dimension of variation at the declared tolerance — and its product is a determinate, one specific way of occupying that dimension; determination demand is thereby a measure over unresolved determinables, the verdict variable's support being the determinate-space at the declared grain where the predicate closes. | -| `DDD-frame-14` | conceptual | projected | Discharge always produces a determinate, which lands in two registers: as an outcome — the determinate as it lands in the world, produced at every completed act — and as a verdict — the determinate as assessed by a declared predicate, produced only where governance has declared one; governance is thereby the conversion of outcomes into verdicts, and every diachronic instrument runs on verdicts alone. | -| `DDD-frame-15` | conceptual | projected | At every completed act in a task's scope, the act's determination demand is discharged — by a filed decision, an actor's judgment, an arrangement default, or an uncontrolled draw; escape is a supply mode of discharge, not an absence of supply — demand is never unmet, only ungoverned. | -| `DDD-frame-16` | conceptual | projected | Discharge is act-indexed: standing supply is inherited per act and occasioned supply is produced per act, so there is no act-free discharge — governance never chooses whether demand is supplied, only by what, chosen in advance or defaulted at the act. | -| `DDD-ground-01` | normative | projected | A governing decision must declare a resolvable applicability predicate, unless it explicitly declares universal applicability; non-evaluation must never silently become non-applicability, and where the predicate is implemented over declared ground axes, each axis is marked mechanically-evaluable or judgement-evaluable. | -| `DDD-ground-02` | conceptual | projected | Source coverage (covered · declared-empty · undeclared · unknown), resolution (resolved · deliberately-open · unknown), and assurance (adequate · inadequate · unknown) are orthogonal properties of ground relative to a filed decision set: only source coverage = undeclared is a source-coverage finding, deliberately-open is a resolution value carrying a deferred verdict, and Unknown is never a pass. | -| `DDD-ground-03` | conceptual | projected | A decision whose resolution is deliberately-open has no resolution for a timing predicate to read, so the timing vocabulary carries a fourth value — "—(open)" — alongside before, during, and after; any timing predicate without it misfiles open decisions, because every definite value it can assign erases the declaredness that makes them open. | -| `DDD-ground-05` | conceptual | projected | Declaring the determinable space is constitutively prior to determination over it — a determination selects a determinate, and determinates exist only as ways of occupying a declared determinable — with the symmetry that ground is prior within each act while decisions are prior in a registry's growth, so the priority is synchronic constitution and the bootstrap diachronic history, not a circle. | -| `DDD-hyp-01` | empirical | projected | Holding generation difficulty and resources constant, comparative advantage shifts toward high-throughput computational generators as acceptance becomes more operationally evaluable, feedback becomes faster and denser, ground becomes more accessible, checking becomes cheaper, and retries become more affordable. | -| `DDD-hyp-02` | empirical | projected | Human or institutionally situated arrangements retain greater comparative advantage as relevant ground becomes unavailable to the computational system, consequences are delayed, evaluators disagree, acceptance criteria drift over time, tacit or socially distributed knowledge is required, or normative legitimacy is part of the task. | -| `DDD-hyp-03` | empirical | projected | A generator-plus-checker arrangement — a computational generator composed with a mechanical checker or an independent reviewer — outperforms both generator-alone and judgment-alone baselines where candidate generation benefits from breadth or speed, significant parts of acceptance are operationally closed, and the remaining open residue can be escalated. | -| `DDD-hyp-04` | empirical | projected | Trust and deployment willingness are better predicted by the completeness of the accountability arrangement — attribution, persistent principal, authority linkage, stake, sanction path — than by whether the immediate executor is human or computational. | -| `DDD-hyp-05` | empirical | projected | Reliance on worker or provider selection increases as result-level evaluation becomes slower, less objective, less stationary, and less complete, holding labour supply, training cost, consequence severity, and task structure constant. | -| `DDD-measure-01` | empirical | projected | Specification demand is verdict entropy: for a task whose acceptance predicate closes, the demand engineers experience as specification burden is H(V) over the ground distribution. | -| `DDD-measure-02` | formal | established | Given the identification (DDD-measure-01), conservation on the closing region is the chain rule of entropy: H(V) = I(V;X) + H(V\|X) for any conditioning variable X. | -| `DDD-measure-03` | formal | reported | The seam of a decomposition is I(V;S); a decomposition with cheaper parts has pre-paid more demand into the seam, and H(V\|S) is minimised exactly when I(V;S) is maximised. | -| `DDD-measure-06` | formal | established | The measure exists iff the acceptance predicate operationally closes; H(V) is undefined exactly where the framework's floor result locates non-zero floor. | -| `DDD-measure-08` | formal | retired | RETIRED — "a better decomposition destroys demand." Cheaper parts were purchased by a higher-information seam; the destruction was an artifact of not counting I(V;S). | -| `DDD-measure-10` | formal | established | You cannot decompose your way out of the work: for a fixed closing task, H(V\|S) = 0 requires I(V;S) = H(V) — the parts become trivial only when the decomposition already encodes the entire verdict. Demand is conserved, not escapable by re-decomposition. | -| `DDD-measure-11` | conceptual | reported | The measure prices the verdict, not the search: H(verdict) is a property of the verdict function and the ground distribution and says nothing about the cost of computing a correct answer. Two tasks with identical verdict entropy can differ unboundedly in generation cost, so the measure must not be read as pricing generation. | -| `DDD-measure-15` | conceptual | projected | The engineering reading of the chain-rule identification holds only for admissible conditioning variables. A conditioning variable X is admissible where it is computable from ground available at the act and from what the arrangement has standing before it, and not from the verdict itself — computable by something that has not been handed the answer. The arithmetic holds for any X whatever; admissibility is what restricts the reading, not the identity. | - -### Decisions - -| ID | Statement | -|---|---| -| `DDD-dec-15` | The escape mechanism's scope correction: overflow ∩ open is the mechanism of capacity-generated escape — sufficient for escape, never necessary for it. Escape as such stays as term:escape defines it, supplied by nobody for any reason, and term:escape-mechanism and DDD-floor-01 are re-scoped by supersession of their universal quantifier. | -| `DDD-dec-24` | Wave 3 files the principle layer's central material: the indexed-determination core given its canon home (core/14, with the flag clearances that home licenses), the discharge chain (the determinable, the determinate's two registers, supply-mode exhaustiveness, act-indexed discharge, constitutive priority of ground), and the hypothesis set as a new claim area — six derivation-grade claims, five statement-grade hypotheses, three settled terms, one new core document, filed by supersession-and-amendment with every divergence from source material reported rather than harmonised. | -| `DDD-dec-26` | The Track 1 session mints two terms and no more — term:commitment-level and term:residual-discretion, both established by core/14-indexed-determination.md §2 — discharging the mint DDD-frame-02 deferred pending use; three anticipated deltas (axis, arrangement, escaped decision) are ruled non-deltas because canon already carries each, and the five-way ground-provenance taxonomy is ruled ineligible for minting because its institutional slot is Q27-gated. | - -### Terms - -| ID | Term | Canonical wording | -|---|---|---| -| `term:accountability` | accountability | **Accountability** is a property of the arrangement, not of the executor: attribution of the determination, a persistent answerable party, and a borne consequence. An arrangement missing any of the three has not allocated the decision's consequence. | -| `term:actor` | actor | An **actor** is a system that resolves decisions by reading ground: variation in declared ground can alter the resolution through an internal pathway that selects among alternatives. A thermostat qualifies; a falling rock does not. Actorhood does not require intelligence. | -| `term:admission-test` | admission-test | **A choice is a decision iff varying *the choice* moves the outcome past tolerance.** **A fact is ground iff varying *the world* moves the outcome past tolerance.** | -| `term:answerability` | answerability | **Answerability** — the obligation to produce the chain: which determinations were made, by whom, against what ground. | -| `term:arrangement` | arrangement | The **arrangement** is the composition through which a resolution is produced and governed: executor, prior commitments, ground channels, checks, reviewers, record, and accountable principal. The unit of comparison is the arrangement, not the isolated actor. | -| `term:attribution` | attribution | **Attribution** — provenance-shaped, and therefore checkable: the record connecting the determination to the execution that produced it. | -| `term:closure` | closure | **Effective closure, defined.** A predicate is **closed for an arrangement** when the relevant ground is observable and adequacy can be evaluated within declared resource, latency, and confidence bounds. **Decidable** is reserved for the formal special case. | -| `term:commitment-level` | commitment-level | A **commitment level** is a level at which an arrangement fixes behaviour in advance: **outcome-level** — permitted resolutions fixed directly; **policy-level** — the generating procedure fixed; **principal-level** — a determiner selected by qualification and case-level resolution delegated. The three compose, and they are levels of commitment, not species of actor: the question is never which of three kinds an actor is, but at which levels the arrangement has committed. | -| `term:composite-actor` | composite-actor | **A composite actor carries its members' demand, plus the seam demand between them.** | -| `term:conservation` | conservation | **For a task at a declared assurance level, and within a fixed decomposition of that task, determination demand is conserved.** Every governing decision gets made. The only choice is **by whom, when, and at what price.** Reduce the demand in one store and it **relocates**; it does not vanish. | -| `term:determinable` | determinable | The **determinable** — an outcome-relevant dimension of variation at the declared tolerance: the object determination resolves, and the dimension of comparability an axis names. Determinateness comes in orders — red → scarlet → this shade — and the declared tolerance names the order at which the framework stops distinguishing. | -| `term:determinate` | determinate | The **determinate** — one specific way of occupying a determinable: what discharge produces. A determinate is a way of being, not the determinable plus a differentia, and determinates under one determinable are constitutively exclusive at their grain. | -| `term:ensemble-actor` | ensemble-actor | **The choice is a property of the ensemble, and it exists nowhere in any member.** | -| `term:escape` | escape | **Escaped** — determined *never*, by nobody: decided-by-nobody as a first-class category. Latent defect exposure. **The only forbidden state.** | -| `term:escape-mechanism` | escape-mechanism | **Capacity-generated escape — the escape an actor produces from residual it has taken up — requires two conditions, both necessary:** **(1) Overflow** — demand exceeds resolve capacity. **(2) Open** — no verifier the actor holds. Overflow alone (closing predicate) → **retries, not escape.** Recoverable. Not floor. Open alone (within capacity) → **carried by judgment**, where an accountable supplier is named. Not floor. Where none is named, it is escape by another route (`05` §7) — outside this mechanism, not excluded by it. **Overflow AND open** → **escape. This is the floor.** **Sufficient for escape, never necessary.** A governing decision that never entered an actor's residual escapes without overflowing anything: escape is supplied-by-nobody (`term:escape`), and capacity shortfall is one generator of it. | -| `term:floor` | floor | The "floor" is the portion of a determination's demand that **cannot be moved off the in-the-moment actor** — the residue that no amount of encoding or checking can amortise, that must be paid, per run, in judgment. **The intrinsic floor is a property of the acceptance predicate, not of the decision.** | -| `term:liability` | liability | **Liability** — bearing the consequence. | -| `term:outcome` | outcome | The **outcome** — the determinate as it lands in the world, produced at every completed act. The **verdict** is the same determinate as assessed by a declared predicate, produced only where governance has declared one. The world renders outcomes, never verdicts; governance is the conversion of outcomes into verdicts. | -| `term:residual-discretion` | residual-discretion | **Residual discretion** is the outcome-relevant variation remaining at the act after the arrangement's declared commitments are applied. It is not randomness: a deterministic arrangement can carry substantial discretion across unfamiliar cases, a randomised one can be tightly committed, and a zero-variance arrangement can be consistently wrong. | -| `term:seam` | seam | A composite carries the demand of its parts, **plus** the seam demand `S` created *between* them. | -| `term:store` | store | **{rule, check, actor, nothing}.** There is no fifth source. | -| `term:swarm-gate` | swarm-gate | **A swarm is an actor only if it genuinely determines choices against ground.** The admission tests (`00` §4) still gate, and they must. | -| `term:training` | training | **Closure decides whether training is *available*. Cost decides the *ratio* when it is.** Training requires a **reliable error signal** — you must be able to tell, and tell soon, whether the output was right. | -| `term:verdict` | verdict | **Definition (determination demand).** *(In the engineering projection this same quantity is denominated in the vocabulary of the domain and called* **specification demand** *; the measure below is identical either way.)* For a task whose acceptance predicate **closes** for the arrangement (`term:closure`; *decidable* is the formal special case, not the requirement), the predicate evaluates outcomes, and the **task class** supplies one correct output per input point. The **verdict** is that induced assignment — the correct output over each point of the input space. Let `P` be the distribution over inputs (the *ground distribution*). The **determination demand** of the task is the Shannon entropy of the verdict: **D = H(verdict)**, measured in **bits**. Where the task class supplies no such assignment, the predicate still evaluates outcomes and there is no verdict to have entropy about — which is the boundary `09` §7 draws. | - -### The hypothesis set, as the graph holds it - -The hypothesis set is broken out because its discipline is the easiest thing in the paper for a -reader to mistake. Every row is `projected`, every row declares a falsifier, **every evidence field -is empty**, and every row is owned by a study that has not been run. The columns below are the -graph's own fields, not the paper's summary of them. - -| ID | Status | Evidence | Owner | Falsifier declared | -|---|---|---|---|---| -| `DDD-frame-07` | projected | `[]` (empty) | paper-4 | yes | -| `DDD-hyp-01` | projected | `[]` (empty) | paper-4 | yes | -| `DDD-hyp-02` | projected | `[]` (empty) | paper-4 | yes | -| `DDD-hyp-03` | projected | `[]` (empty) | paper-4 | yes | -| `DDD-hyp-04` | projected | `[]` (empty) | paper-4 | yes | -| `DDD-hyp-05` | projected | `[]` (empty) | paper-4 | yes | - -*Generated from the graph at `v5.9.0`. 45 claims, 3 decisions, 24 terms.* From c186ee805d9f18eefb2a1f7d7e0f200e101ebe0f Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 31 Aug 2026 13:51:11 +0000 Subject: [PATCH 7/9] GATE 4 follow-up: the measure note's commit pin, resolved to the real commit MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The supplement's S2 carried a placeholder while the note's W1 edits were uncommitted — a hash cannot be cited before it exists. Now resolved to the commit this session created, with the reason stated: Paper A pinned the note at aa7e135, W1 edited the note, so the pin advances to a commit this revision produced rather than citing prose that says ground distribution while this paper says deployment distribution. Basis: DDD-dec-34; DDD-delivery-02; term:delivery --- papers/paper-a/paper-a-supplement.md | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/papers/paper-a/paper-a-supplement.md b/papers/paper-a/paper-a-supplement.md index e5649c8..48e93e5 100644 --- a/papers/paper-a/paper-a-supplement.md +++ b/papers/paper-a/paper-a-supplement.md @@ -36,8 +36,11 @@ it in one step. Two of the four quotations that broke at the advance are quotati review asked the paper to stop overstating, so the repair paid a debt the paper already owed. The companion measure note — `papers/measure-note/measure-note.md` in `decision-driven-design` — is -cited by path at commit `d6bc1a1`, which carries its discharge section and the vocabulary rename -described in S5. The note is absent from `v0.4.0` and from every earlier downstream tag, so it +cited by path at commit `2b26f25`, which carries its discharge section and the vocabulary rename +described in S5. **That commit is this session's own** — Paper A pinned the note at `aa7e135`, and +W1 edited the note, so the pin had to advance to a commit this revision created. The alternative was +to keep citing a commit whose prose says `ground distribution` while this paper's says +`deployment distribution`, which would have been a citation that reads as current and is not. The note is absent from `v0.4.0` and from every earlier downstream tag, so it resolves at a commit rather than at a tag. --- From 15435dbc76f26d798be56501e2ca283b34948cf9 Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 31 Aug 2026 14:03:30 +0000 Subject: [PATCH 8/9] =?UTF-8?q?GATE=205:=20close=20=E2=80=94=20the=20respo?= =?UTF-8?q?nse=20to=20review,=20the=20manifest,=20and=20what=20the=20full?= =?UTF-8?q?=20read=20caught?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit THE FULL READ CAUGHT TWO REAL DEFECTS, both invisible to all three checkers because neither is a block quotation, an appendix row, or a status assertion. 1. The ABSTRACT still carried the retired four-mode enumeration — "by a filed decision, an actor's judgment, an arrangement default, or an uncontrolled draw" — after §4.1 had been repaired to DDD-frame-17's three values. The paper's most-read paragraph was projecting a claim retired at v5.10. Fixed. 2. Two references to "Appendix A" survived in a paper whose appendix had moved to the supplement. Fixed. Recorded rather than quietly corrected because they are the argument for the full read: three green checkers and a clean validator run did not see either, and the second checker gap this session found is the same shape as the first. ALSO from the read: §3.3's trained-model case now works the review's own counterexample explicitly — policy-committed on one axis, predetermined on another, standing on a third, all true at once and none competing. The old list read as competition because it drew values from three axes at once. And the abstract's opening names WHICH accounts held the determiner fixed rather than implying all of them did, which the survey no longer supports. response-to-review.md maps all twenty-two objections: twenty-one conceded, repaired, or accepted and booked; ONE defended, a wording choice whose underlying finding is conceded. The headline was corrected during drafting — an earlier version claimed three defences by counting propositions the review PRAISED, which would have flattered us. The manifest records the close checks, the word count with its method and the growth attributed, the pin advance's prediction against its observation, the four gates' rulings, and three findings that are not repairs: the unpinned node that fired nothing, the corroboration check firing on a claim that would have HELPED the paper, and the charter's 88% that was really 35%. Successor item 4 carries Emil's GATE 4 rule: an advance without instruments is unobservable, and an unobservable advance is presumed discharge at the pin layer. Close checks, all green: 30 verbatim / 0 failing · 96 status assertions / 0 wrong · 75 nodes / 0 discrepancies · validate-core-order 0 errors 0 warnings, 71 pins 0/0/1 · 26 claims with the 6 recorded warnings · 22 decisions · 22 references with 0 cited-but-absent and 0 absent-but-cited · 58 headings with 0 dangling section references. Body 11,708 -> 17,334 words (+48%), supplement 5,232, by the method that reproduces the review's own figures to within 1%. actor-indexed-determination is untouched, which is the correct state for a projection session. Basis: DDD-dec-20; DDD-dec-34; DDD-agent-01; DDD-frame-15; DDD-frame-17; DDD-frame-16; DDD-frame-08; DDD-measure-06; DDD-measure-16; DDD-cost-20; term:commitment-level; term:accountability; term:residual-discretion; term:delivery --- .../2026-08-30-paper-a-revision/manifest.md | 140 +++++++ meta/successor-items-paper-a-revision.md | 15 +- papers/paper-a/paper-a.md | 35 +- papers/paper-a/response-to-review.md | 363 ++++++++++++++++++ 4 files changed, 541 insertions(+), 12 deletions(-) create mode 100644 meta/sessions/2026-08-30-paper-a-revision/manifest.md create mode 100644 papers/paper-a/response-to-review.md diff --git a/meta/sessions/2026-08-30-paper-a-revision/manifest.md b/meta/sessions/2026-08-30-paper-a-revision/manifest.md new file mode 100644 index 0000000..c939417 --- /dev/null +++ b/meta/sessions/2026-08-30-paper-a-revision/manifest.md @@ -0,0 +1,140 @@ +# Manifest — Paper A's revision, carrying W1 (2026-08-30) + +**Session type:** interactive paper revision, projection work. Five gates, held at every one. +**Principal:** Emil. **Branch:** `claude/paper-a-w1-revision-b8sx45`, both repositories. +**Base:** `actor-indexed-determination` `81f6929` (= `v5.12.0`, verified) · `decision-driven-design` +`54f00eb`. + +**Nothing merged. No claim, term or decision was filed, amended or retired in either repository by +this session**, except the pin-advance decision this repository owes for its own pin. Q44, Q45 and +the ontology explainer were read and cited nowhere. + +--- + +## 1. What the session produced + +| Item | Result | +|---|---| +| **R-1** the related-work survey | seven works engaged, every locator verified against Crossref, Open Library or NCBI; three entries record a neighbour being stronger than the framework | +| **R-2** narrowed claim, retitle, rhetoric | *The Missing Parameter* → *Actor-Indexed Determination*; the claim narrowed in the review's own words; five warrant-overrunning formulations repaired | +| **R-3** pin advance and quotations | `v5.9.0` → `v5.12.0`, predicted before it ran and verified after; four quotations repaired, and a fifth repair no checker could see | +| **R-4** the supplement split | `paper-a-supplement.md`; the apparatus moved, not reduced, and still running | +| **R-5** W1 | 15 moved, 16 deferred, of 31 in the two manuscripts | +| **R-6** the review's small items | §7's arity mapped, §8.1's provenance made hypotheses, §8.5's escapes made candidates, §9.5's Study 0 added, and — flagged rather than assumed — the closure ladder repaired | + +## 2. The close checks + +| Gate | Result | +|---|---| +| `check-quotations.py` @ `v5.12.0` | **30 verbatim, 0 disclosed-partial, 0 failing** | +| `check-status.py` @ `v5.12.0` | **96 inline status assertions, 0 wrong** | +| `check-appendix.py` @ `v5.12.0` | **75 nodes rendered, 75 cited, 0 discrepancies** | +| `gen-appendix.py` idempotence | byte-identical across three runs, in split mode, manuscript unwritten | +| `validate-core-order.py core/` | **0 errors, 0 warnings**; 71 pins, 0 basis-loss, 0 content-drift, 1 standing shadow | +| `validate-claims.py core/claims/` | valid, 26 claims, **6 warnings** — the six `CLAUDE.md` records at `v5.11.0` | +| `validate-claims.py core/decisions/ --decisions` | valid, 22 decisions | +| reference closure | 22 entries; **0 cited-but-absent, 0 absent-but-cited** | +| internal cross-references | 58 headings, **0 `§` references matching no heading**; 0 references to material that moved to the supplement | + +## 3. Word count, with the method stated and the growth attributed + +**Method:** regex word tokens over the Markdown source, `[A-Za-z0-9][A-Za-z0-9'’\-–—/.]*`. It +reproduces the external review's own figures — *"approximately 11,600 words before Appendix A; +14,900 in total"* — to within 1%, which is why it is the method used rather than a better one. + +| | Arrival (`40d277f`) | Close | +|---|---|---| +| Argumentative body | **11,708** | **17,334** | +| Supplement | — | **5,232** | +| Together | 15,250 | 22,566 | +| Body headings | 63 | 70 | + +**The paper grew 48%, and the attribution is the defence.** The review asked for narrower claims, a +survey it did not have, and less machinery in the argumentative path. All three were done. The +machinery left — 5,232 words of it. What replaced it is qualification: a narrower claim needs the +hedging a broad one does not; a survey of seven works is longer than an assertion about them; three +of its entries record a neighbour being stronger, and each costs a paragraph; and a status +discussion that says what `established` does *not* mean is longer than one that says what it does. + +**No argument was cut** (Emil, GATE 3). The failure mode this session was written to avoid is a +shorter paper that sounds more certain because it has less to qualify, and buying brevity out of +argument would have been that failure with extra steps. + +## 4. The pin advance, predicted and verified + +Committed at `7c9e2fa` **before** `graph/upstream.yaml` was touched, with the instrument that +computed it beside it. Advanced in two stages — ref first, hashes second — so the firing could be +observed rather than assumed. + +| | Predicted | Observed | +|---|---|---| +| `E12` | 0 | **0** | +| `W5` | exactly 1 — `DDD-frame-15`, projected → retired | **1, that one** | +| `W6` | exactly 6, named by id, field and both hashes | **6, every id and every hash as written** | +| `W7` | 1, unchanged | **1** | + +After re-instrumentation: 71 pins, 0 basis-loss, 0 content-drift, 1 shadow. Filed as `DDD-dec-34`. +The migration seed's own three-W6 prediction for this advance appeared exactly; none of its seven +W1 W6 fired, correctly, since W1 renames nothing upstream in this session. + +## 5. Three findings that are not repairs + +**5.1 A node forcing the largest repair in the revision fired nothing.** `DDD-measure-06` retires at +`v5.12.0` from `established`. `W5` sees only pinned ids and it was not pinned; both existing +checkers passed all three prose lines calling it `established`, because none is a block quotation +and none is an appendix row. Found by a sweep written at GATE 1, now shipped as +`papers/paper-a/check-status.py`. Four nodes pinned in consequence. + +**5.2 The corroboration check fired on a claim that would have helped us.** A search result +attributed a *"Law of Conservation of Complexity"* to Woods & Hollnagel's *Patterns*. Had it held it +would have given the framework's own conservation principle a named precedent inside the literature +the review says we ignore. **It could not be corroborated in any primary source and is used +nowhere.** A verification rule that only ever removes inconvenient claims is not a verification +rule; this is the case worth having on record. + +**5.3 A count in the session's own charter did not reconcile.** The charter said 88% of W1 lives in +the two papers. Against the committed classification, **88 is the corpus-wide mutable total and the +two manuscripts hold 31 of it — 35%.** The migration seed is corrected, and W1's remaining surface +is **73 of 88, not zero.** + +## 6. Rulings taken at the gates + +| Gate | Ruling | +|---|---| +| 1 | `status-sweep.py` becomes a third checker · retitle candidate 1 · survey plan approved, the DOI error to go to the response · supplement boundary as stated · no argument cut · **all five ambiguous occurrences defer, including the three reading as S5** · stale line numbers to the migration seed's method rule | +| 2 | §11 ratified, including the three where-a-neighbour-is-stronger entries and the Bovens finding at its true size · reading grades in reference entries become standing practice · length accepted with attribution · the conservation-law near-miss to the manifest as a finding | +| 3 | ladder repair ratified rather than reverted · §7's arity: the paper flags **and** a canon session gets the issue · rhetoric pass not over-corrected · `README.md:32` to the freight list | +| 4 | the measure note's pin follows the note · the checkers point at the note **before** its pin next advances, because *an advance without instruments is unobservable, and an unobservable advance is presumed discharge at the pin layer* · §5.4's heading renames with the divergence clause | + +## 7. Files + +**`decision-driven-design`** + +| Path | Change | +|---|---| +| `papers/paper-a/paper-a.md` | retitled; §4.1, §3.2, §3.3, §4.4, §5.1, §5.2, §6.1, §7, §8.1, §8.5, §9.5, §11, §12 revised; the appendix, the pin apparatus and the reproduction notes removed to the supplement | +| `papers/paper-a/paper-a-supplement.md` | **new** — the reproduced nodes, the pin, the filing statuses, the arity finding, the W1 record, Reproduction | +| `papers/paper-a/check-status.py` | **new** — the third checker | +| `papers/paper-a/gen-appendix.py`, `check-appendix.py` | a fourth argument: citations read from the manuscript, rendered into the supplement; new defect history in both | +| `papers/paper-a/response-to-review.md` | **new** — every objection mapped | +| `papers/paper-a/reviewer-brief.md` | retitled | +| `papers/measure-note/measure-note.md` | W1: 14 moves, 16 deferrals; §5.4's asset-divergence clause | +| `papers/measure-note/measure-paper-context.md` | Paper A's title, with the reason | +| `graph/upstream.yaml` | ref `v5.9.0` → `v5.12.0`; six pins re-instrumented; four added | +| `core/decisions/DDD-dec-34.yaml` | **new** — the pin advance | +| `meta/migration-plan-ground.md` | the third method mechanism; W1's counts corrected | +| `meta/successor-items-paper-a-revision.md` | **new** — six items | +| `meta/way-of-working.md` | Paper A's title | +| `meta/sessions/2026-08-30-paper-a-revision/` | prompt, bootstrap, four gate documents, three instruments, package README | + +**`actor-indexed-determination`** — **untouched.** The principle repository carries no reference to +its dependents, and this session had no reason to write to it. Its branch exists and is empty of +changes, which is the correct state for a projection session. + +## 8. What is owed after this + +Six successor items at `meta/successor-items-paper-a-revision.md`: the accountability arity, W1's +remaining 73 occurrences, the `measure-nonuniform-ground.py` divergence, the measure note's missing +checkers, `README.md:32`'s over-claim on the freight list, and the unverifiable conservation-law +attribution. The response-to-review's §8 lists what the *paper* still owes, which is a different and +longer list, and the first item on it is the measure's decoder repair. diff --git a/meta/successor-items-paper-a-revision.md b/meta/successor-items-paper-a-revision.md index 3fcbe57..13cdfb8 100644 --- a/meta/successor-items-paper-a-revision.md +++ b/meta/successor-items-paper-a-revision.md @@ -98,8 +98,19 @@ protected by one session's reading and by nothing else, and a later session rena would notice. The note is also three releases behind on its own pin. The three scripts are already ref-parameterised and manuscript-parameterised, so pointing them at -the note is a small session rather than a new instrument. **It should happen before the note's pin -advances**, not after: the checkers are what make an advance observable. +the note is a small session rather than a new instrument. + +**Ruled at GATE 4 (Emil): the checkers point at the note before the note's pin next advances**, and +the reason is a rule rather than a preference for this note: + +> **An advance without instruments is unobservable, and an unobservable advance is presumed +> discharge at the pin layer.** + +That is the same shape as `term:presumed-discharge` one layer up: the artefact recording a verified +advance is identical to the artefact recording an unverified one, because a green run and no run +produce the same file. The two-stage verification this session used — advance the ref, observe the +firing, then re-instrument — requires instruments to observe with. Without them the note's next pin +advance is a bump, and a bump is what the pin discipline exists to prevent. --- diff --git a/papers/paper-a/paper-a.md b/papers/paper-a/paper-a.md index c938532..cc77569 100644 --- a/papers/paper-a/paper-a.md +++ b/papers/paper-a/paper-a.md @@ -16,10 +16,13 @@ argument.* ## Abstract -Several influential accounts of complexity, control and verification describe where the work of -determination resides while holding the determiner largely fixed. That omission has become +Several influential accounts of complexity allocation — Ashby's regulator, Brooks's essential and +accidental complexity, Tesler's conservation, Meyer's contracts — describe where the work of +determination resides while holding the determiner largely fixed. Holding it fixed has become consequential, because engineered systems now combine authored procedures, learned policies, human -judgment, mechanical checks and institutional principals inside one arrangement. +judgment, mechanical checks and institutional principals inside one arrangement. **Other +literatures did not hold it fixed** (§11), and this paper's contribution is stated against them +rather than against their absence. We state **actor-indexed determination**: unresolved determination is relative not to a task alone but to a tuple of task, accessible ground, acceptance relation, declared tolerance, system @@ -30,10 +33,10 @@ consequential resolution that no adequate source-and-assurance combination gover unacceptable outcomes are not reliably detected. Two propositions give the framework its shape, and both are **proposals with declared falsifiers -rather than findings**. Every completed act's determination demand is discharged — by a filed -decision, an actor's judgment, an arrangement default, or an uncontrolled draw — so **demand is -never unmet, only ungoverned**; and discharge is act-indexed, so governance never chooses whether -demand is supplied, only by what. On the region where the acceptance predicate closes, demand is +rather than findings**. At every completed act, each outcome-relevant alternative is discharged in +exactly one of three ways — predetermined, exercised, or drawn — so **demand is never unmet, only +ungoverned**; and discharge is act-indexed, so governance never chooses whether demand is supplied, +only by what. On the region where the acceptance predicate closes, demand is identified with the entropy of the verdict, and on that identification conservation is the chain rule of entropy. The identification is a modelling claim and it is unproven; the chain rule is Shannon's and is certain and empty on its own. **The construction has three conditions, not one** — @@ -387,6 +390,17 @@ answers are not — with the important qualification that what training buys is changes the arrangement's commitments without changing the model at all, which is §2.5's point arriving in a concrete form. +**This is the case the external review used to break the paper's earlier discharge list, and it is +worth showing the repair on it.** A trained model was, on that list, policy commitment, judgment +and standing encoded supply at once — three classifications from one deployment, which is a +partition failing. The repair is that the three were never on one axis. On **commitment level** the +model is policy-committed [term:commitment-level, draft]. On **discharge** it is predetermined, +where the standing configuration and the ground jointly fix the output, and exercised or drawn where +they do not — which depends on the decode and not on the weights (§4.1) [DDD-frame-17, projected]. +On **act-indexation** its supply is standing, inherited per act [DDD-frame-16, projected]. All three +readings are true simultaneously and none competes with another, because they answer different +questions. The earlier list read as competition because it drew values from all three axes at once. + *An escalation path.* An arrangement that commits at outcome level for familiar cases and delegates unfamiliar ones to a named principal has committed at two levels simultaneously. This is the ordinary case in deployed systems, and the composition clause is what lets the framework describe @@ -964,7 +978,7 @@ The graph carries **two counts** of accountability's elements, and the external that the paper used both without supplying the mapping. The claim above names five — attribution, persistent principal, authority linkage, stake, sanction path. The canonical term names three: *"the attribution of the determination, a persistent answerable party, and a borne consequence"* -[term:accountability, settled; reproduced in full in Appendix A]. The mapping is not one-to-one and +[term:accountability, settled; reproduced in full in the supplement]. The mapping is not one-to-one and the difference is not cosmetic. | `term:accountability` (three, **settled**) | `DDD-frame-08` (five, **projected**) | Relation | @@ -1149,8 +1163,9 @@ The five hypotheses below are **predictions, not findings**. Every one is `proje graph, every one carries a declared falsifier, **every one's evidence field is empty**, and every one is owned by a study that has not been run. The emptiness is discipline, not oversight: the framework files a hypothesis at the strength its falsifier gives it and does not backfill evidence -from the corpus that suggested it. Appendix A shows the status and the empty evidence column for -all six nodes so that no reader has to take this paragraph's word for it. +from the corpus that suggested it. The supplement reproduces the status and the empty evidence +column for all six nodes, straight from the graph, so that no reader has to take this paragraph's +word for it. They replace binary human-versus-model predictions with graded predictions about **arrangements** (§2.5). Each names a gradient, not a boundary. diff --git a/papers/paper-a/response-to-review.md b/papers/paper-a/response-to-review.md new file mode 100644 index 0000000..16f82f3 --- /dev/null +++ b/papers/paper-a/response-to-review.md @@ -0,0 +1,363 @@ +# Response to the external review of *Actor-Indexed Determination* + +**The review:** `meta/sessions/2026-08-23-phase1a/paper-a-objective-review.md`, against the +manuscript at `40d277f` and the graph at `v5.8.0`. Recommendation: **major revision / reject and +invite resubmission.** + +**This response:** the revision of 2026-08-30, session record at +`meta/sessions/2026-08-30-paper-a-revision/`. Every objection the review raises is mapped below to +**conceded**, **repaired**, or **defended — and why**. Nothing is left unanswered, including the +objections the paper lost. + +**The headline, stated first because it is the honest summary.** We count **twenty-two distinct +objections** in the review. **Twenty-one are conceded, repaired, or accepted and booked as work not +yet done. One is defended** — a wording choice, and even there the finding underneath it is +conceded. §5 below lists three things we decline to give up, but those are propositions the review +*praised*, not objections it raised, and counting them as defences would flatter us. + +Two of the three load-bearing findings had already forced **canon** changes before this revision +ran, so much of what the paper does here is catch up with corrections your review caused. **The +paper is longer than the one you reviewed, not shorter** — 11,708 words to 17,334, a 48% increase — +and §9 explains why we think that is the honest outcome rather than a failure to comply. + +--- + +## 1. Two corrections to the review itself + +Offered because they are useful to you, not to score points. + +**1.1 The Hollnagel and Woods citation points at the wrong volume.** §8 cites +`10.1201/9781420005684` as *"Woods and Hollnagel, Joint Cognitive Systems"*. That DOI resolves to +**Woods & Hollnagel (2006), *Joint Cognitive Systems: Patterns in Cognitive Systems Engineering***. +The volume your prose is about — the joint cognitive system stated as a unit of analysis — is +**Hollnagel & Woods (2005), *Foundations of Cognitive Systems Engineering***, +`10.1201/9781420038194`. Two books, one year apart, sharing a main title, with the author order +reversed. **Both are now cited**, since both are on point, and the correction is reported rather +than made silently because a database would have propagated the error. + +**1.2 A near-miss on our side, recorded for symmetry.** While drafting the survey we nearly +attributed a *"Law of Conservation of Complexity"* to *Patterns* — a claim that, if true, would +have **strengthened** our position by giving our conservation principle a named precedent in the +literature you say we ignore. It could not be corroborated in any primary source. **It is used +nowhere**, and it is recorded in our successor list so a later session does not rediscover and +believe it. We mention it because a verification rule that only ever removes inconvenient claims is +not a verification rule. + +--- + +## 2. The three critical findings + +### 2.1 The closure–measure biconditional is false (review §2) — **CONCEDED, and repaired in canon** + +You are right, and the worse fact is that the contradiction was ours before it was yours. Our own +companion measure note had already rewritten its §7 into a scope condition with three separated +requirements — existence, availability, estimability — and **canon was never updated to match**. So +`DDD-measure-06` still carried the biconditional at `established`, and the paper projected it +faithfully. **The projection was stronger than the note sharing its graph.** + +- `DDD-measure-06` is **retired** (`retired_from: established`), superseded by `DDD-measure-16` + (availability to an arrangement, at `established`) and `DDD-measure-17` (the coincidence, at + `projected` — see 2.1b). +- The paper now separates the three conditions in a table at §4.4 and carries the separation at + five further sites. **Closure governs availability and neither of the other two.** +- Your four-condition repair (§2.5) is adopted in substance. We fold semantic determinacy and + constructive closure differently — the first into existence, the second onto our closure ladder — + and say so. + +**2.1b One thing your finding produced that you did not ask for.** The old node bundled two +propositions under one status: *where the measure applies*, and *what the agreement between that +boundary and the floor's location is worth*. The second is an interpretive claim about two +arguments, not arithmetic, and it is now `projected` rather than `established` — with a falsifier. +Splitting them was your finding's consequence and it is a better claim than either half was. + +**2.1c And a finding about our instruments, not our claims.** `DDD-measure-06` retired and **fired +no warning**: our cross-repository pin instrument only watches nodes that are pinned, and this one +was not. Both existing checkers passed all three lines where the paper called it `established`, +because none is a block quotation and none is an appendix row. We found it with a sweep written for +this revision, now shipped as a **third checker** (§7 below). The node and its successors are now +pinned. + +### 2.2 Raw ground collapses the allocation measure (review §3) — **CONCEDED; interim posture adopted** + +Take `X = G` for a deterministic task and mutual information reports the ground as carrying the +entire verdict, while the arrangement may hold no rule for turning that ground into an answer. +`DDD-measure-15`'s admissibility condition does not exclude it. **You are right, and the previous +round's answer — a caveat that a real actor may fail to use information that is present — does not +repair a condition that fails to exclude the case.** + +**What we have not done:** repaired it. The framework-native fix is that admissible `X` should be +*a representation the arrangement can decode*, which would make the residual arrangement-relative +and bring the measure into line with the paper it appears in. That is research, not filing, and it +is not in this revision. + +**What we have done:** adopted your own recommended interim posture. The paper does not present +verdict entropy as an established measured region. §4.4 states the identification as a modelling +claim, `projected`, with its falsifier a correspondence that has not been run; §10.2 states the +admissibility condition's limit; and the abstract says the identification is unproven in the same +sentence that says the chain rule is certain. + +### 2.3 The four discharge modes do not partition (review §4) — **CONCEDED; the claim is retired** + +All three of your overlaps land. `DDD-frame-15` is **retired**, superseded by `DDD-frame-17`. + +You asked for either an explicit priority rule that assigns each case exactly once, or a recast onto +orthogonal axes. **A priority rule was drafted and tested against your six cases. It ties on +declared defaults and on a timeout's when-to-stop alternative, and every ordering that breaks those +ties crosses the guard separating governance-supply from discharge.** So the disposition is the +recast, not the rule — and your framing of the choice is what produced that answer. + +The successor partitions by two dichotomies and nothing else: does the standing configuration +together with the ground at the act fix the resolution (**predetermined**), and if not, is what +fixes it inside the arrangement's control (**exercised**) or outside it (**drawn**). Three +consequences, each closing one of your overlaps: + +- **the unit is the outcome-relevant alternative, not the act.** Your timeout case dissolves here: + a timeout carries a when-to-stop alternative and a what-to-return alternative and they discharge + differently. The retired claim quantified over acts in its statement and over alternatives in its + own test field — a second defect you did not report and we found while drafting the successor; +- **governance status is not an axis.** Your declared-default overlap was a declaredness test + smuggled into a locus partition. Declaredness is a different partition's question; +- **no value turns on actorhood.** The retired *judgment* mode did, and thereby inherited the + admission test's circularity you name in §6. This one does not inherit it. + +Your trained-inference case is worked explicitly at §3.3: policy-committed on one axis, +predetermined on another, standing on a third, **all three true at once and none competing**. The +old list read as competition because it drew values from three axes simultaneously. + +**One thing we defend.** You suggest *"ungoverned resolution"* would be clearer than *"determined +never"*. We keep the canonical phrase, because it names a state and *"ungoverned"* names an +assessment — and we accept your underlying finding: an external reader met our guard and still +reported the phrasing as conflicting, which is a finding about our exposition. §4.1 now states the +guard before the reader can trip on it rather than after. + +--- + +## 3. The major findings + +### 3.1 Residual discretion conflates different phenomena (review §5) — **CONCEDED; the term is amended** + +Your cryptographic-hash example is decisive and canon now carries it. `term:residual-discretion` is +amended to quantify **at fixed ground**: the alternatives still admissible once the standing +configuration *and this act's ground* are both given. A hash varies enormously across inputs and +exercises no discretion, because at each input the commitments fix the output exactly. Under the old +wording it came out carrying enormous discretion, which was plainly wrong. + +Your four phenomena are separated: variation across ground, epistemic uncertainty about a fixed +policy, stochasticity, and genuine unresolved selection. Three of the four are now excluded in the +term's own text. `DDD-frame-02`'s clause is amended in step. + +**A consequence worth naming:** this repair and 2.3's turn out to be one repair made in two nodes. +An alternative with no residual discretion at fixed ground is exactly what *predetermined* means. + +### 3.2 The actor admission test remains partly circular (review §6) — **CONCEDED, and left open** + +You are right that "selection" is not independently defined, and that the rock's failure is produced +by the intended meaning of the word rather than by a test. **We have not repaired it and we do not +claim to have.** It is flagged as open rather than patched, because a patch that redefined selection +without an independent criterion would move the circularity rather than remove it. + +What we can report is that the discharge partition **no longer depends on it** — its three values +turn on determination and control, and are answerable without deciding whether a producer is an +actor (2.3). The circularity is contained rather than cured. + +### 3.3 Split relevant ground from accessible ground (review §6) — **ACCEPTED, not yet executed** + +`G*` (relevant world facts) versus `G_A` (ground accessible and delivered to arrangement `A`) is +correct and strengthening, and it is the delivery distinction applied to ground. It is **canon work +we have not done**: the tuple's `ground` coordinate is a settled term, and a projection may not +redefine one. The paper carries the distinction informally where it bears — §8.1 now says that +ground missing to one arrangement is available to another that can run the search, and that the +floor moves with it — and the split is booked. + +### 3.4 Graph status is being used as epistemic status (review §7) — **CONCEDED; partly repaired** + +This is the objection with the widest consequence and we treat it as such. + +**Repaired in the paper.** The note on claim status now says explicitly that `settled` and +`established` mean argued and unchallenged *inside this framework* and **not externally validated**; +that `reported` means a computation reproduces, not that the world was consulted. The conclusion +states that **every `established` claim in this graph is `formal`** — so what is settled here is +arithmetic — and that **nothing in either list has been checked against the world**. Your phrase for +the analysis sections, *authorial synthesis not represented in the graph*, is adopted in substance: +`(analysis)` now says it carries no status *"which does not make it uncontestable, only unattributed +to the graph."* + +**Repaired in the instruments.** A third checker verifies that every inline status assertion asserts +the status the graph carries (§7 below). + +**Not repaired: the kind/maturity split.** You are right that `projected` bundles definitions, +formal claims, empirical hypotheses and normative prescriptions under one maturity label. The graph +carries a `kind` field, which the paper's tables now render — but the split you actually ask for +runs *through* several claims rather than between them, and a session that examined it ruled that +assigning a finer value before those claims are split would miscode whichever job it did not name. +**It is sequenced, not dismissed**, and it is the change we most want to land before these +repositories become public. + +**Your point about the primary claim's falsifier is conceded** and §9.5 now says it: failure of the +umbrella prediction would not falsify the ontology, because the framework's descriptive and +predictive claims are separately falsifiable. + +### 3.5 Novelty is not established (review §8) — **CONCEDED; the claim and the title are withdrawn** + +The largest single piece of work in this revision. + +- **The title is withdrawn.** *The Missing Parameter* asserted an absence we had not surveyed. The + paper is now *Actor-Indexed Determination*, and §1.2 says why the old title went. +- **The narrowed claim is adopted in your words**: a specific, auditable synthesis of resolution, + assurance, delivery and accountability — not the discovery that sociotechnical arrangements + matter. +- **All seven works are engaged**, each with what the framework takes and where it differs, with a + comparison table. Every locator is verified against Crossref, Open Library or NCBI. + +**Three entries record where a neighbour is stronger than the framework**, because a survey that +only found advantages would not have been a survey: + +- **Horvitz supplies a decision-theoretic criterion for allocation and we supply none** off the + closing region; +- **Leveson's control structure carries authority structurally** and we do not improve on it; +- **four of `DDD-frame-08`'s five accountability elements are already Bovens's.** What remains is + authority linkage and a change of tense from retrospective assessment to design-time constitution. + One element and a tense, not a new theory. + +**One difference claim was withdrawn by the reading itself.** Our survey plan had said Matthias's +responsibility gap reads as an arrangement naming an executor and no principal — *"fixable rather +than novel"*. The primary text does not support *fixable*: the gap rests on a **control condition on +just ascription**, not on a missing name. The paper now says structural completeness does not close +it, and that whether it is morally sufficient is a normative question our claim is `projected` about. + +**What survived the survey**, and it is what we now claim: no work surveyed carries +filing-versus-delivery as a standing category, and none separates a resolution's source from its +assurance mechanism as orthogonal dimensions. The paper adds, in its own words, that **a distinction +being unoccupied is not a distinction being correct.** + +### 3.6 Hypotheses are not study-ready (review §9) — **CONCEDED; coding reliability comes first** + +§9.5 now names **Study 0**: inter-rater reliability and predictive validity of the framework's +coding scheme, **before any comparative work**, with agreement reported *per dimension rather than +pooled* — since a scheme can be reliable on closure and unreliable on discharge, and a pooled figure +would hide exactly that. + +Two admissions follow, both yours: + +- **the escaped-decision count is circular** until the instrument holds. A count produced by the + framework's own criterion cannot also be evidence for it, and it is usable only as an inter-rater + quantity until an independent criterion exists — which we do not have; +- **H1 and H2 bundle variables**, difficulty and resources are invoked as controls without + operational definitions, and each hypothesis is a research programme rather than a study. The + paper says so. + +--- + +## 4. The smaller findings + +| Review | Disposition | +|---|---| +| **§10.1** the closure ladder mixes axes; decidability is logical, not a stronger operational rung | **Conceded and repaired.** The ladder is now **three rungs, all operational**, ending at constructively closed. Decidability is stated as the logical axis's answer, with both directions of non-implication given: a decidable predicate need not be operationally closed, and an operationally closed one need not be decidable. The defect was in our framing, not in canon, which always reserved *decidable* for the formal special case | +| **§10.2** accountability's competing element counts | **Conceded and mapped.** §7 supplies the table: the five-element version *refines* the third element into stake and sanction path and **adds** authority linkage. It refines-and-adds; it does not replace. **And a `projected` claim that adds an element to a `settled` term is a supersession question**, which the paper reports and does not rule — a projection may not rule on canon. It is filed as a successor item so the flag reaches canon rather than sitting where nothing points back at it | +| **§10.3** the worked example's provenance classifications | **Conceded and restated.** Every assignment in §8.1 is now a hypothesis with *what would overturn it*. Your schema challenge is the instructive one: an arrangement that both maintains a schema and reads it from a store holds it under **two provenances at once**, and our five-way partition does not adjudicate. That is stated as a limit | +| **§10.3** the two claimed escapes may be discoverable | **Conceded.** They are now **candidate** escapes, and our own vocabulary supplies the objection: an escape is a property of an arrangement, so a walk that did not look everywhere has not established one. Code search, contract tests, telemetry and asking the consuming teams all reach the question the walk assumed missing | +| **§11** length, repository-nativeness, graph machinery in the argumentative path | **Conceded and executed.** The generated tables, the pin mechanics, the filing history and the reproduction instructions are in a separate supplement. The boundary rule: *a passage moves when it is about the graph; it stays when it is about the world, or about how much warrant the paper has* | +| **§11** the rhetoric overruns the status labels | **Conceded and repaired.** "Two results give the framework its shape" is now "two proposals with declared falsifiers rather than findings"; "its strongest result" is "its most load-bearing proposal — and a proposal"; the conclusion no longer asserts the framework's value but states its aim, and names the open question | + +--- + +## 5. What we decline to give up + +**These are not defences against objections.** All three are propositions the review identifies as +worth keeping, and we list them because a revision that conceded this much needs to say what it did +*not* trade away under pressure. + +**5.0 The one actual defence** is smaller than any of them, and it is in 2.3: we keep the canonical +phrase *"determined never, by nobody"* rather than replacing it with *"ungoverned resolution"*, +because the first names a state and the second names an assessment. Your finding underneath the +suggestion — that a reader met our guard and was still confused — is conceded, and §4.1 is +restructured because of it. + +**5.1 The arrangement as the unit of analysis.** You call it the durable core and we agree — but we +defend *keeping* it while conceding the priority. The novelty narrows to the synthesis; the object +stays. + +**5.2 Filing versus delivery.** You called it a strong engineering contribution unprompted, and the +survey supports it: **no work we read carries it as a standing category.** Leveson's causal-scenario +step can reach the case, and meaningful human control's tracing condition is adjacent to it — an +arrangement can satisfy tracing completely, with a named and answerable designer, and still fail at +the act because the decision that designer authored was never retrieved. **A tracing audit passes +that.** The category is the contribution and we keep it. + +**5.3 The disclosure discipline.** You note twice that the absence of evidence is correctly +disclosed. We have not traded it for brevity: **every projected-and-empty sentence stayed in the +main paper** when the apparatus moved to the supplement, on the explicit ground that those sentences +are about warrant rather than about the graph. + +--- + +## 6. Where the revision leaves the paper by your own scorecard + +| Dimension | Your assessment | What changed | +|---|---|---| +| Formal correctness | *critical boundary error* | the node is retired; three conditions separated; the successor's coincidence claim demoted to `projected` | +| Conceptual consistency | *several load-bearing conflicts* | the discharge partition recast; residual discretion amended; the closure ladder repaired; the accountability counts mapped | +| Related work | *far below publication standard* | seven works engaged, locators verified, three of them recording where the neighbour is stronger | +| Originality | *novelty not yet demonstrated* | the claim is narrowed to your formulation and the title withdrawn | +| Empirical support | *none, correctly disclosed* | unchanged, and Study 0 now precedes the comparative design | +| Writing | *overlong and over-defensive* | **the paper is longer.** See §12 | +| Internal traceability | *excellent* | a third checker added, covering the surface the other two structurally cannot reach | + +--- + +## 7. One thing the review caused that is not in the paper + +Your §7 is about status being read as warrant. Acting on it exposed a gap in our instruments rather +than in our prose. + +We had two checkers: one verifying block quotations against the graph, one re-reading the generated +tables. Neither reads the manuscript's **most common** citation form — the inline citation asserting +a status in running prose. When `DDD-measure-06` retired, the paper called it `established` in three +places and **both checkers passed all three lines.** + +`check-status.py` now verifies that surface, and it ships with its origin recorded as its warrant: +it was not designed against a specification, it was generalised from a sweep that found something, +**so its coverage is that sweep's and no wider.** That the one unchecked surface was exactly the one +converting a citation into warrant is not a coincidence we wanted to leave unfixed. + +--- + +## 8. What is still owed + +Stated because a response listing only what it fixed is not a response. + +1. **The measure's decoder repair.** `I(V;X)` is not usable-information and we have not made it one. + The interim posture is adopted; the repair is research. +2. **`G*` versus `G_A`.** Accepted, canon work, booked. +3. **The kind/maturity split.** Sequenced behind a claim-splitting question it depends on. +4. **The actor admission test's circularity.** Open, and flagged as open. +5. **The accountability arity.** Reported by the paper, filed as a successor item, unruled. +6. **Inter-rater reliability itself.** Study 0 is designed and unrun. Every hypothesis's evidence + field is still empty, and the paper still says so on every page it needs to. + +--- + +## 9. Recommendation, in your own terms + +Your §13 judged the manuscript strong as a canonical repository statement, promising for a workshop +after major revision, and reject-and-resubmit as an archival conceptual paper. We think the revision +moves the third of those and do not claim it clears it — **the measure's construct-validity problem +is not repaired, and you were right that it is load-bearing.** What the paper now does is state that +limit where a reader meets the measure, rather than in a footnote. + +**The one thing we would ask you to weigh again** is the relationship between length and +compliance. You asked for narrower claims, better sourcing, and less machinery in the argumentative +path. We did all three, and the argumentative body grew from **11,708 words to 17,334 — 48%** +(counted by the same method that reproduces your own figures of ~11,600 and ~14,900 to within 1%: +regex word tokens over the Markdown source). + +The machinery did leave: 5,232 words of it are now in a supplement. **What replaced it is +qualification.** A narrower claim needs the hedging a broad one does not. A survey of seven works is +longer than an assertion about them. Three of its entries record a neighbour being stronger than we +are, and each of those costs a paragraph. The status discussion now says what `established` does not +mean, which is longer than saying what it does. + +**We would rather defend the length than the certainty that buying brevity would have cost.** The +failure mode we were most worried about entering this revision was a shorter paper that sounds more +confident because it has less to qualify. If the length is now the objection, that is a better +objection to have. From ce54cb6d2fd02f22c686a1390ecdfd6bc255e264 Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 31 Aug 2026 14:08:20 +0000 Subject: [PATCH 9/9] GATE 5 addendum: the uninstrumented-surface finding, and its general form MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit On Emil's GATE 5 ruling: record the abstract finding in the manifest as the argument for the full read, and state the general form because it will recur. THE FORM. Instruments cover the surfaces someone thought to instrument. The abstract is the surface nobody thinks to instrument, because it is prose ABOUT the paper rather than a citation IN it — so it inherits no citation's protection while carrying more of the reader's warrant than any single citation does. This was the second unwatched-surface finding of the session and the same shape as the first: §5.1's was an unpinned node moving silently, this one is unciting prose asserting silently. A third checker answered the first. A fourth is deliberately NOT proposed for this one — an instrument that parsed an abstract for claims it does not cite would guess at meaning rather than check correspondence, and its false negatives would be indistinguishable from a clean run. The three checkers verify correspondence; that is what makes them trustworthy and what bounds them. The remedy is the full read, and its cost is now known. Successor item 7 carries the standing instruction: in any session repairing a projection against a moved graph, the abstract and the conclusion are repaired last and read whole, because a restatement inherits nothing from the citation it paraphrases. Whether it belongs in meta/way-of-working.md as a projection rule is left to whoever next touches that document, and is not filed here. Manifest §5.2 also now records both checks that fired against claims which would have HELPED the paper — the unverifiable conservation law, and this session's own draft headline claiming three defences by counting propositions the review praised. Twice in one session, which is the only evidence that the checks are not decorative. Basis: DDD-dec-20; DDD-agent-01; DDD-delivery-02; DDD-frame-15; DDD-frame-17; term:escape --- .../2026-08-30-paper-a-revision/manifest.md | 41 ++++++++++++++++++- meta/successor-items-paper-a-revision.md | 30 ++++++++++++++ 2 files changed, 69 insertions(+), 2 deletions(-) diff --git a/meta/sessions/2026-08-30-paper-a-revision/manifest.md b/meta/sessions/2026-08-30-paper-a-revision/manifest.md index c939417..f7b6178 100644 --- a/meta/sessions/2026-08-30-paper-a-revision/manifest.md +++ b/meta/sessions/2026-08-30-paper-a-revision/manifest.md @@ -85,14 +85,51 @@ checkers passed all three prose lines calling it `established`, because none is and none is an appendix row. Found by a sweep written at GATE 1, now shipped as `papers/paper-a/check-status.py`. Four nodes pinned in consequence. -**5.2 The corroboration check fired on a claim that would have helped us.** A search result +**5.2 Two checks fired against claims that would have helped us, and that is the pattern worth +keeping.** The first is the corroboration check. A search result attributed a *"Law of Conservation of Complexity"* to Woods & Hollnagel's *Patterns*. Had it held it would have given the framework's own conservation principle a named precedent inside the literature the review says we ignore. **It could not be corroborated in any primary source and is used nowhere.** A verification rule that only ever removes inconvenient claims is not a verification rule; this is the case worth having on record. -**5.3 A count in the session's own charter did not reconcile.** The charter said 88% of W1 lives in +The second fired on this session's own writing. A draft of `response-to-review.md` opened +*"nineteen conceded or repaired, three defended"* — a count reached by treating three propositions +the review **praised** as objections we had defended. It would have flattered us. Corrected during +drafting to twenty-one conceded and **one** defended, with §5 renamed *What we decline to give up* +and told plainly that those are not defences. **Twice in one session a check fired against something +that would have helped**, which is the only evidence that the checks are not decorative. + +**5.3 The abstract carried a retired claim past three green checkers, and this is the argument for +the full read.** At GATE 5, with `check-quotations.py`, `check-status.py` and `check-appendix.py` +all reporting zero and every repository validator clean, the abstract still read *"discharged — by a +filed decision, an actor's judgment, an arrangement default, or an uncontrolled draw"*. That +enumeration was retired at `v5.10` and §4.1 had already been repaired to carry its successor. **The +paper's most-read paragraph was projecting a claim the graph had killed**, and nothing caught it but +reading the paper from the top. + +*Why no instrument saw it.* The abstract is not a block quotation, so the quotation checker does not +read it. It carries no citation, so the status checker has nothing to compare. It is not an appendix +row. **Three instruments, three surfaces, and the sentence most readers will actually read sits +outside all of them.** + +**The general form, stated because it will recur:** + +> **Instruments cover the surfaces someone thought to instrument.** The abstract is the surface +> nobody thinks to instrument, because it is *prose about the paper* rather than a citation *in* it +> — so it inherits no citation's protection while carrying more of the reader's warrant than any +> single citation does. + +This is the **second** unwatched-surface finding of this session and it is the same shape as the +first. §5.1's was an unpinned node moving silently; this one is unciting prose asserting silently. +Both were found by a human-scale pass rather than by a checker, and both were found only because +someone went looking at a layer the instruments do not describe. A third checker was the answer to +the first. **A fourth is not obviously the answer to this one** — an instrument that parsed the +abstract for claims it does not cite would be guessing at meaning rather than checking +correspondence, which is a different kind of tool and a worse one. The honest remedy is the full +read, and its cost is now known. + +**5.4 A count in the session's own charter did not reconcile.** The charter said 88% of W1 lives in the two papers. Against the committed classification, **88 is the corpus-wide mutable total and the two manuscripts hold 31 of it — 35%.** The migration seed is corrected, and W1's remaining surface is **73 of 88, not zero.** diff --git a/meta/successor-items-paper-a-revision.md b/meta/successor-items-paper-a-revision.md index 13cdfb8..c110d80 100644 --- a/meta/successor-items-paper-a-revision.md +++ b/meta/successor-items-paper-a-revision.md @@ -148,3 +148,33 @@ Hoffman & Woods's *Beyond Simon's Slice*, and records no conservation law from * volume itself could not be obtained. **Nothing in the paper rests on it.** Anyone with the book in hand can settle it in a minute, and it is a real finding either way. + +--- + +## 7. The abstract is an uninstrumented surface, and the finding generalises + +**Recorded at GATE 5 (Emil): note the general form, because it will recur.** + +At this session's close — three checkers green, every validator clean — the abstract still carried +the four-mode enumeration retired at `v5.10`, after §4.1 had been repaired to carry its successor. +Nothing caught it but reading the paper from the top. + +> **Instruments cover the surfaces someone thought to instrument.** The abstract is the surface +> nobody thinks to instrument, because it is *prose about the paper* rather than a citation *in* it +> — so it inherits no citation's protection while carrying more of the reader's warrant than any +> single citation does. + +**What this item is not.** It is not a proposal for a fourth checker. An instrument that parsed an +abstract for claims it does not cite would be guessing at meaning rather than checking +correspondence — a different kind of tool, and a worse one, since its false negatives would be +indistinguishable from a clean run. The three checkers all verify correspondence against the graph; +that is what makes them trustworthy and it is also what bounds them. + +**What it is.** A standing instruction for any session that repairs a projection against a moved +graph: **the abstract and the conclusion are repaired last and read whole**, because they restate +claims without citing them, and a restatement inherits nothing from the citation it paraphrases. +The full manifest entry is at `meta/sessions/2026-08-30-paper-a-revision/manifest.md` §5.3, with the +sibling finding — an unpinned node moving silently — at §5.1. + +Whether this belongs in `meta/way-of-working.md` as a projection rule rather than sitting in a +session's successor list is a question for whoever next touches that document. It is not filed here.