Populate passive_partnership_s_corp_income: SCF non-active-business shape, Form 8960 level

PROGRESS.md

@@ -1,65 +1,67 @@
-# Progress: one US target surface
+# Progress: passive-pass-through-722 re-verdict closure
 
 ## State
 
-Surface unification and scale-identity enforcement are complete on
-`one-target-surface`: every US entrypoint
-now compiles one national + state + congressional-district target registry,
-and the CLI/config switches that could delete CD or JCT target rows are gone.
-Parity doctrine now protects that family as a red-line compile. Sparse and dense
-artifacts may differ in record count, never target membership.
+Re-verdict findings B1 and B3 are closed locally offline on
+`passive-pass-through-722` from starting HEAD `bc0ddb0a`. No production code,
+restricted artifacts, or supervisor-owned tests have been changed. No fetch,
+push, pull-request mutation, rebase, reset, checkout, or stash is in scope.
 
 ## Done
 
-- Read `CLAUDE.md`, the target-parity declaration, the fiscal compiler and its
-  never-controls doctrine, the current CD opt-in tests, and
-  [microcosm#449](https://github.com/PolicyEngine/microcosm/issues/449) /
-  [microcosm#569](https://github.com/PolicyEngine/microcosm/issues/569).
-- Attempted the required `uv sync --all-packages --extra us`. The managed
-  sandbox denied writes to the default uv cache, then its network restriction
-  prevented a clean-cache download of `pyvis`. A byte-identical-lock sibling
-  environment was cloned copy-on-write; tests use that complete environment
-  with this worktree's package sources first on `PYTHONPATH` because an offline
-  editable reinstall still requires unavailable build-isolation metadata.
-- Attempted the GitNexus refactoring impact workflow. Local indexing completed,
-  but GitNexus could not register the index because the sandbox forbids writing
-  `~/.gitnexus/registry.json`; a direct source/call-site audit is the fallback.
-- Confirmed the starting worktree was clean and no build or push was run.
-- Ran the workspace suite for 1,137 seconds with no failure before interrupting
-  it inside the unrelated PUF-QRF stale-checkpoint subprocess regression; the
-  affected US target/compiler shard is the per-commit validation boundary, with
-  a complete workspace run reserved for the final tree. Ruff is green.
-- Established a green 10-file affected-suite baseline covering target
-  compilation, parity, the release builder/scorers, CD vintage translation,
-  Ledger profiles, and the generated calibration contract (100% in 349.59s).
-- Removed the congressional-district compilation option throughout the fiscal
-  compiler, builder, fiscal scorer, state-file scorer, ACS local tool, aging
-  diff, experiments, tests, docs, and generated contract. The canonical CD
-  crosswalk is now the default at each production entrypoint.
-- Removed the diagnostic JCT target-deletion option and its release-gate bypass;
-  diagnostic tools now score the same registry as releases.
-- Removed the parity generator's CD regime switch and regenerated the pinned
-  manifest to 32 compiled / 52 reviewed families. The generated calibration
-  contract declares all three geography layers and has no default-layer split.
-- The 10-file affected suite reaches 100% with exit 0 after the runtime change;
-  Ruff, byte compilation, and `git diff --check` pass.
-- Promoted the CD family into the parity anti-rot red-line set, pinned the
-  manifest's 32/52 header counts to parsed family counts, and asserted that its
-  compiled entry carries no exclusion fields or fence.
-- Strengthened the fiscal invariant: the CD aggregate is present in the
-  compiled registry while the taxable-interest rebase still refuses it as a
-  national control. The standard 10-file affected suite reaches 100% with exit
-  0 after the parity-doctrine change.
-- Removed the compiler's per-run support-exclusion parameter, the release CLI
-  and loader that populated it, its provenance branch, both experiment callers,
-  and the obsolete sparse Build-J exclusion JSON.
-- Added exact signature locks for the compiler, release builder, fiscal scorer,
-  and state scorer; legacy membership flags are parser-rejected.
-- Added a CD-bearing registry identity test across nominal 57,240-record sparse
-  and 337,704-record dense artifacts. Specs and content-addressed registry
-  version are identical. The standard affected suite reaches 100% with exit 0.
+- Read `CLAUDE.md` and the adjudicated B1/B3 sections of
+  `sol-run-reverdict-0822.log` before changing the branch.
+- Read the prior audit's stream-isolation note: its uncommitted historical
+  composition produced byte-identical v1/v2/v3 outputs and preserved global
+  NumPy RNG state.
+- Confirmed the worktree starts clean at `bc0ddb0a` on
+  `passive-pass-through-722`.
+- Attempted the GitNexus exploration workflow. No GitNexus repository resource
+  or query tool is exposed in this session, so direct source and call-path
+  inspection is the fallback.
+- Reserved
+  `packages/microcosm-build/tests/test_spec_engine_engine_abi.py` and
+  `packages/microcosm-build/tests/test_us_multispine_pool_tool.py` for their
+  other owner; this task will not touch them.
+- Froze the exact historical `qbi-v3-wiring` simulator, post-QRF reconciler,
+  and v1/v2/v3 assumption resources from `cfbf6330` as SHA-pinned test-only
+  fixtures. Current HEAD intentionally carries none of that retired runtime.
+- Added the CI-executable synthetic B1 regression over the historical runtime's
+  literal supported-version gate `(1, 2, 3)`. Every parameter realizes a
+  positive passive row count and weighted aggregate, byte-preserves the full
+  15-leaf simulation surface plus its routing-owned self-employment leaf, and
+  preserves both global NumPy state and every version-family draw (including
+  the v2/v3 host-classifier seed).
+- The targeted B1 node completed with all three parameters passing (exit 0).
+  Touched-file Ruff check and format check are clean. An earlier duplicate was
+  manually interrupted during the installed-engine collection import (exit
+  130); it produced no test result. No run exited 137.
+- Replaced the circular B3 calibration check with test-local evidence parsing,
+  inverse-CDF integration, odds shifting, row-wise weighted aggregation, and
+  bisection. Those helpers consume only committed JSON and caller-supplied raw
+  arrays; they call no production calibration or subordinate helper.
+- The two-row synthetic CI node has the hand solution `shift = ln(3)` and
+  aggregate `(100*1 + 200*2) * 0.5 * 0.75 = 187.5`; the test-local and
+  production solvers agree within the artifact-derived binary64 floor. Its
+  final run passed (exit 0).
+- The required restricted replay passed (exit 0) with the supplied SHA-pinned
+  `puf_2024.h5`: test-local solved shift `-1.157105426398319`, committed shift
+  `-1.157105426398319`, effective shift tolerance
+  `2.220446049250313e-16`, and independently seeded aggregate
+  `$55,021,131,518.061035` (`0.7187449327011208%` above the provisional
+  `$54,628,492,000` midpoint). No test run exited 137.
+- Committed the B1 closure as `bd537c42` and the B3 closure as `92dd3568`.
+  Both received independent static review with no defects found.
+- Re-ran the committed B3 synthetic node (`1 passed`) and exercised the
+  env-unset restricted node (`1 skipped`). The earlier artifact-backed replay
+  passed once. Touched-file Ruff check and format check are clean.
+- Updated `FINAL_REPORT.md` and `PROGRESS_PASSIVE_722.md` with a dated
+  2026-08-22 record that separates CI coverage from the env-gated replay,
+  maps each finding to its closing commit, and marks older self-certification
+  language as superseded.
 
 ## Next
 
-- Quantify the target-row delta and existing 25% timing evidence, run final
-  verification, then write `FINAL_REPORT.md`.
+- No local implementation work remains. The supervisor can push the committed
+  handoff and let clean-runner CI arbitrate the suite; no local push or PR
+  mutation is in scope.

PROGRESS_PASSIVE_722.md

@@ -0,0 +1,170 @@
+# Progress: microcosm #722
+
+## State
+
+The adjudicated B1 and B3 re-verdict findings are closed locally on
+`passive-pass-through-722` as of 2026-08-22. B1 closes in `bd537c42`; B3 closes
+in `92dd3568`. The earlier merge and remediation history below remains the
+branch record. Work remains offline: no fetch, push, PR mutation, publication,
+or promotion is in scope, and clean-runner CI remains the suite arbiter.
+
+## Done
+
+- Read the binding adversarial verdict before inspecting or changing the
+  branch.
+- Re-read `CLAUDE.md` and confirmed the PR-CI/certification and root-journal
+  contracts for this remediation pass.
+- Attempted the GitNexus debugging workflow; no GitNexus query/context tools
+  are configured in this session, so direct source and call-path analysis is
+  the documented fallback.
+- Confirmed the branch, detached pre-approved merge worktree, merge base, and
+  current `origin/main` identities without network access.
+- Resolved the only three merge conflicts with the approved typed country
+  resource rows, both independent multispine imports, and one measured runtime
+  graph pin. The three resolved file blobs match the pre-approved detached
+  index exactly.
+- Ran the post-merge source-blindness graph test itself with repository
+  conftests disabled because the pre-sync environment does not yet contain
+  main's new `jsonschema` dependency. The single registered tool reaches the
+  pinned 67-module graph (base 64, branch +2, main +1).
+- Traced the country-package closure failure to
+  `tools/generate_us_bundle_from_constants.py::LEGACY_RESOURCE_PATHS`; the
+  loader's `_assert_file_closure` compares that emitted manifest with package
+  files.
+- Committed the real merge of current `origin/main` as `d9f19c66`; its second
+  parent is `2c7a7218` and it is not a rebase or squash.
+- Declared the three passive/repeal JSON resources on
+  `LEGACY_RESOURCE_PATHS`, completed the full generator run, and ran the exact
+  offline `uv sync --all-packages --all-extras` equivalent against the locked
+  environment.
+- Traced the remaining constants-adapter failure to the passive contract being
+  present only as generated pool code. Added it to the typed pipeline schema,
+  compiler identity contract, and stacked-authority projection, then regenerated
+  the bundle at spec SHA-256
+  `11b5f830141d007b53131f6c863d15a6be10787fc6d88aeb147317d2c1f593e0`.
+- Independently reviewed the changed field-usage claims: 62 new pointers total
+  (29 authored, 33 resolved), all accounted for by the three typed manifest
+  rows and the typed passive pipeline/pool-code contract.
+- Regenerated the F0 evidence report with exact 41,442/41,442 field coverage
+  and 40/40 inventory checks. The targeted adapter, schema/identity,
+  field-ledger, inventory, and coverage suite passes 45 tests; the US generator
+  check and targeted Ruff checks are also clean.
+- Replaced the vacuous passive wrapper preservation proxy with current-main QBI
+  reconciliation on both the baseline and passive-staged paths. Seed 13
+  realizes exactly two nonzero rows (positions 25 and 36) while all 15 incumbent
+  QBI leaves retain identical dtype and bytes.
+- Strengthened the production multispine regression with a $2 million
+  partnership-income fixture. Production seed 0 realizes exactly one nonzero
+  row (position 5, amount `158609.82342210703`) and every incumbent QBI leaf is
+  byte-identical to the no-passive current-main pipeline. The independent
+  full-length RNG-family array test remains in place; all three focused tests
+  pass and targeted Ruff is clean.
+- Made `us_release_input_coverage_gate` require every hard-required manifest
+  column to exist in `engine.variables()`, independently of frame presence and
+  signal. Failures name the version read from `engine_abi.lock.json` and report
+  `required_missing_from_engine_registry` in gate details.
+- Added the two explicit failure-mode regressions
+  `test_required_all_zero_column_fails_when_registry_contains_it` and
+  `test_required_column_absent_from_locked_registry_fails_even_with_signal`.
+  The former proves default-only data stays red even with registry presence;
+  the latter proves nonzero data stays red when the locked variable is absent.
+- Documented in the generated release manifest, passive assumptions, repeal
+  benchmark resource, and US fact-to-target guide that the passive column is
+  engine-inert under PolicyEngine-US 1.764.6 and the hard release gate remains
+  red until the pin advances past 1.764.6 to a release containing #9306.
+- Regenerated the manifest and restricted assumptions, then propagated their
+  identity through the US bundle (spec SHA-256
+  `00b4c73b0ff2e29abf00b9f6f8112c9b87e937d323fb5c973bb5cd0652a95931`)
+  and F0 report. The targeted gate/resource/inventory/adapter suite passes 153
+  tests in 8m34s; the generator check and targeted Ruff are clean.
+- The 2026-08-21 restricted replay check called the production calibration
+  solver and therefore did not constitute an independent solve. Its former
+  certification claim is superseded by the 2026-08-22 B3 entry below. The
+  ordinary assumptions validator change and its historical test results remain
+  valid branch history.
+- The implementation and verification bullets below describe the historical
+  pre-review checkpoint at `a4d93f78`; the verdict supersedes its former
+  completion claim.
+
+- Read `CLAUDE.md` and established the PR-CI/certification and journal
+  contracts.
+- Confirmed the primary checkout has unrelated untracked paths and left it
+  untouched.
+- Created the requested isolated worktree at
+  `.claude/worktrees/passive-722` from current `origin/main`.
+- Attempted the GitNexus exploration workflow; no GitNexus MCP resources or
+  tools are configured, so source and Git-history inspection are the fallback.
+- Confirmed the old QBI v3 stack is opt-in, diverges before the repository
+  rename, and does not expose its latent entity form on the output frame.
+- Selected a sibling passive assignment stage before current QBI
+  reconciliation. This preserves the archived 15-leaf QBI contract and all
+  prior QBI random streams while still accepting a latent form for routing
+  when one becomes available.
+- Reproduced the six SCF Schedule-E-band cells from the local 2022 public
+  extract. Presence cells all clear effective n=30; the three middle-band
+  conditional-share cells fall back to the pooled holder sample.
+- Confirmed the restricted PUF replay artifact is present and pinned its
+  size, digest, row counts, weighted positive pass-through aggregate, and
+  provisional Form 8960 midpoint calibration target.
+- Added the reproducible provisional SCF evidence resource. Holding
+  prevalence rises from 3.35% in the nonpositive Schedule-E band to 38.98%
+  above $1 million; conditional-share cells in the three thin middle bands
+  use the documented all-holder fallback.
+- Added the version-1 sibling assignment and assumptions build. The persisted
+  log-odds shift is `-1.157105426398319`; its expected aggregate is the
+  $54.628492 billion midpoint and the seed-0 replay produces $55.021132
+  billion (0.72% high, inside the 5% diagnostic tolerance).
+- Verified the restricted replay (12 focused tests), deterministic assumptions
+  regeneration, strict resource hashes, isolated PCG64 families, latent-form
+  routing, and byte preservation of all 15 existing QBI leaves.
+- Integrated the sibling between Schedule-D completion and QBI reconciliation
+  in multispine and direct support builds, while preserving already-assigned
+  rebuilt support. The 997-row remaining-stage audit and support/pool
+  checkpoint identities now bind the evidence, assumptions, shift, and RNG.
+- Added the new output to ownership, L0 export, and release-coverage contracts.
+  The exact pending-engine exception is limited to the input arriving in
+  PolicyEngine-US PR #9306 and becomes a no-op once that engine version lands.
+- Added separate, diagnostics-only Form 8960 line 12 and line 4c aggregate
+  rows. They cannot affect the release gate and retain row-local error
+  containment under the currently locked pre-#9306 engine.
+- Completed the exact final-tree workspace suite: 6,394 passed and 73 skipped
+  in 42m45s. The restricted artifact replay separately passed all 12 tests.
+- Completed repository-wide `ruff check .`, changed-file Ruff format checks,
+  `git diff --check`, the spec-only/entrypoint guards, and a manual incumbent
+  package-name sweep over the branch diff. The manual sweep is required because
+  the ordinary tree guard deliberately excludes `.claude/` worktrees.
+- Added and committed the towncrier fragment and wrote the completion report
+  to `FINAL_REPORT.md`.
+
+## 2026-08-22 — B1/B3 re-verdict closure
+
+- **B1 maps to `bd537c42`.** A secrets-free CI test freezes SHA-pinned
+  historical v1/v2/v3 simulators, reconciler, and assumptions from `cfbf6330`.
+  Its three parameters each require positive passive realization, guard and
+  byte-compare the complete literal 15-leaf QBI simulation surface with and
+  without the passive stage, compare the routed self-employment result,
+  preserve global NumPy RNG state, and compare all version-family draws. The
+  targeted node passed all three parameters.
+- **B3 maps to `92dd3568`.** The secrets-free CI test uses test-local JSON
+  parsing, inverse-CDF integration, an odds transform, row-wise weighted
+  aggregation, and bisection. It reproduces the hand-computable `log(3)` shift
+  and `$187.50` target, then cross-checks the production solver; the targeted
+  node passed once. These test-local routines import or call no production
+  calibration helper.
+- The B3 real replay remains env-gated. Ordinary CI collects it but explicitly
+  skips when `POPULACE_PUF_2024_H5` is unset (`1 skipped` was exercised); a set
+  but missing path raises `FileNotFoundError`. With the supplied restricted
+  artifact it passed once: independently solved and committed shifts are both
+  `-1.157105426398319`; the effective binary64 shift tolerance derived from
+  the JSON's 128-iteration bracket is `2.220446049250313e-16`; the independently
+  reconstructed seeded aggregate is `$55,021,131,518.061035` and its midpoint
+  error is `0.7187449327011208%`.
+- Touched-file Ruff check and format check are clean. No test run exited 137.
+  The earlier full-suite counts above were not rerun during this high-load
+  closure and must not be read as results for the new B1/B3 commits.
+
+## Next
+
+- The supervisor can push the local commits and let clean-runner CI arbitrate
+  the suite. The PolicyEngine-US #9306 dependency and provisional calibration
+  follow-ups remain unchanged; this task performs no promotion.

changelog.d/722-passive-pass-through.added.md

@@ -0,0 +1,3 @@
+Add a provisional SCF-shaped, Form 8960-anchored assignment for the passive
+partnership and S-corporation income NIIT input, with persisted replay
+calibration, release coverage, and diagnostics-only administrative checks.