Skip to content

Prepare SameWrite v1.4.2 (packaging only) - #18

Merged
ipeterpetrus merged 3 commits into
mainfrom
release/v1.4.2-rc
Sep 23, 2026
Merged

Prepare SameWrite v1.4.2 (packaging only)#18
ipeterpetrus merged 3 commits into
mainfrom
release/v1.4.2-rc

Conversation

@ipeterpetrus

@ipeterpetrus ipeterpetrus commented Sep 22, 2026

Copy link
Copy Markdown
Owner 
PACKAGING_ONLY=YES

IMPLEMENTATION_BASE=b966dc126973f3fd113ad02820a2af0df0605f2c

RUNTIME_CODE_DELTA=0
MODEL_FACING_DELTA=0

TARGET_VERSION=1.4.2

PR13_ALREADY_MERGED=YES
PR16_ALREADY_MERGED=YES
ISSUE15_CLOSED=YES

ISSUE14_REMAINS_OPEN=YES
ISSUE17_REMAINS_OPEN=YES

TAG_CREATED=NO
RELEASE_CREATED=NO

Changed files (exactly five)

file change
.claude-plugin/plugin.json version 1.4.1 → 1.4.2
.claude-plugin/marketplace.json metadata.version 1.4.1 → 1.4.2
.codex-plugin/plugin.json version 1.4.1 → 1.4.2
README.md see the README edits below
docs/RELEASE_NOTES_1.4.2.md new

README edits, all active release pointers:

  • The Hermes raw URL now points at tag v1.4.2.
  • The OpenClaw archive URL now points at tag v1.4.2.
  • The README/tag pairing sentence now names v1.4.2.
  • "Intentionally not active in 1.4.2" (it is still true: nothing new is activated).
  • _1.4.2.md is added to the release-notes index.

No descriptions changed, and no file under tools/, skills/, hooks/ or adapters/ changed.

Version-string audit

Before editing, 34 lines matched 1.4.1/1_4_1. Eight were active and were changed:

  • 3 manifests;
  • README lines 16, 17, 27, 28 and 92.

The index line (README line 368) gained _1.4.2.md; its _1.4.1.md entry stays.

The 26 lines that remain are all historical and are kept:

  • README 124, 125, 129. The v1.4.1 accounting correction.
  • README 180, 217. Links to docs/C3_ROUTING_EXPERIMENT_v1_4_1.md (line 129 links the correction document).
  • README 368. The _1.4.1.md entry in the release-notes index.
  • Old release documents. docs/RELEASE_NOTES_1.4.1.md, docs/MEASUREMENT_CORRECTION_1_4_1.md,
    docs/C3_ROUTING_EXPERIMENT_v1_4_1.md and docs/FINDINGS.md.
  • tools/msgid.py comments. Historical measurements; runtime code is not touched.
  • tests/test_multiblock.py:542. A synthetic writer-version fixture.

STALE_ACTIVE_1_4_1_REFERENCES=0.

Evidence

  • Canonical body. CANONICAL_BODY_SHA256=7edec9f21e0bd50583e388e0bdc177f92ba8f0db6ce2bb61acd52b39d81e7767,
    and tools/adapters.py --hashes reports cross-host body identity PASS.
  • Tests. Python 3.10.12: 18 suites, 1708 assertions, 0 failures. tests/test_oneliner_cleanup.sh passes 11/0.
  • Other CI steps. The vnext and presentation selftests, longrun, scan_scale and fixtures_aivos all exit 0.
  • Version-skew selftest. Restoring 1.4.1 in .codex-plugin/plugin.json, in marketplace.json, or in
    the README Hermes pin each turns tests/test_install_paths.py RED (13/1). Restored afterwards.
  • Schema-4 bytes. With the writer-version field held equal, records for a clean, a malformed and a
    bounded fixture are byte-identical between v1.4.1 and this head. The release notes say this, and
    also say the field itself now reads 1.4.2.
  • Overclaim scan. The release notes contain no "fully secure", "all bugs fixed", "best",
    "proven savings" or "autonomous promotion" claims. TOTAL_SAVINGS=NOT_PROVEN and
    WORLD_BEST_CLAIM=NOT_TESTED are stated verbatim.

Not in this PR

No tag, no GitHub Release, no merge. There is no runtime, optimizer, carry or evidence-kernel
change, and no work on #14 or #17. PR #7 is untouched.

This packaging head needs its own independent acceptance before merge, fresh-main CI, the
public install/upgrade validation, the annotated tag and the release.

🤖 Generated with Claude Code

ipeterpetrus and others added 3 commits September 22, 2026 21:02
@ipeterpetrus @claude
Prepare SameWrite v1.4.2 — legacy evidence integrity and candidate-pa…
69bcd98 
…th containment

Packaging only, from main b966dc1. PR #13 and PR #16 are already merged
and independently accepted; this commit changes no runtime file.

* Version 1.4.1 -> 1.4.2 in .claude-plugin/plugin.json,
  .claude-plugin/marketplace.json and .codex-plugin/plugin.json.
* README: the Hermes raw URL and the OpenClaw archive URL pin v1.4.2, the
  pairing sentence names v1.4.2, "Intentionally not active in 1.4.2"
  (still true: nothing new is activated), and the release-notes index
  lists _1.4.2.md. Historical 1.4.1 statements are left as history.
* docs/RELEASE_NOTES_1.4.2.md: what was wrong, what changed in PR #13 and
  PR #16, what did not change, open debt (#14, #17), disclosed residuals,
  compatibility, tests, canonical body identity, install.

Canonical body 7edec9f2…7767 unchanged; tools/, skills/, hooks/ and
adapters/ have zero diff. 18 suites, 1708 assertions, 0 failures on
Python 3.10.12; test_install_paths goes red for a stale manifest or pin.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ipeterpetrus @claude
docs(readme): do not claim a tag-pinned route ran before its tag exists
e0ee3e9 
The install paragraph said every command above "was executed against the
real host ... and is reported only because it worked there". On the
packaging branch those commands pin v1.4.2, a tag that does not exist
yet, so the sentence claimed a result that could not have happened.

It now separates what is true in both states: each install mechanism has
been exercised on its real host in an isolated home; the pinned commands
are checked offline for version and path consistency before publication;
and they run against the published tag during release acceptance, which
can only happen once that tag exists. The OpenClaw cleanup, isolation,
tag-pinning and offline-check statements are kept.

README only. No change to tools/, skills/, hooks/, adapters/, tests/ or
the release notes (they make no such claim).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ipeterpetrus @claude
docs(release-notes): bound the schema-4 compatibility claim, and say …
bf8af6e 
…which rejections cut

Independent packaging review of e0ee3e9 (BLOCK), both findings reproduced
before editing:

* F1 MEDIUM: "for the same input, records are byte-identical to 1.4.1's
  except for the writer-version field" is false. With --max-files and a
  source whose mtime cannot be read (a dangling symlink), the repaired
  bounded_paths() picks a different sample than 1.4.1, so sessions, turns
  and shares differ (5519645a... vs 0c6f05fd...). The notes now separate
  format compatibility (schema 4 unchanged, histories readable) from byte
  identity of new records (not guaranteed; matched-population runs were
  observed to differ only in writer_version, which is an observation).
* F2 LOW: "a rejected line cuts the history" / "a line that failed
  validation cuts every scope". Only rejections classified as loss cut;
  refusals by design, retry bookkeeping and foreign entries do not
  (reproduced: file_global 0 for all three, 1 for a torn line).

Also names the six original defects as R142_01-R142_06 rather than
implying the frozen matrix held only six cases.

Release notes only; README, manifests, tests and runtime untouched.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ipeterpetrus
ipeterpetrus merged commit 43e10cc into main Sep 23, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@ipeterpetrus