Intended as a pull-request body. Neither remote could host one, so it lives in
the repository instead — which is the more durable home anyway, since it travels
with the commits it describes rather than sitting in a forge that may be
re-synced.
- GitHub: the App installation lacks pull_requests:write. POST /pulls returns
403 "Resource not accessible by integration".
- Forgejo: the agent proxy denies core-gitquarters.acbcontent.org at CONNECT
(403, recorded in the proxy's own relay-failure log). Organization network
policy, not a transient failure.
Indexes all 29 commits from 797ebad5 (exclusive) to 199dc72d, each with its
diffstat and an explicit revert classification — Independent, Coupled (with the
group named), or behaviour-changing with the consequence stated. Plus
cherry-pick groupings, three named rollback targets, and the arc's three results.
Records the mirror hazard that already fired once: GitHub's main was reset
backwards from 199dc72d to 797ebad5, which is EXACTLY Forgejo's main. Per
AGENTS.md:280 GitHub is a mirror only, so this was almost certainly a re-sync
from the canonical remote rather than a rewrite. The local clone still had
everything and restored GitHub by fast-forward. Until Forgejo receives this arc,
any further mirror sync will revert GitHub again — which is the single most
important operational fact in this file.
Also records the three hollow mechanisms found in my own work during the arc,
and the two earlier statements that were wrong and were corrected in place
rather than overwritten.
No runtime behavior. One new file.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wcw2pnMBwyvmNyQg4uPEt4
10 KiB
The macro→micro assessment arc — commit index and rollback map
Span: 797ebad5 (exclusive) → 199dc72d (inclusive) · 29 commits · 67 files · +6,340 / −228
Date: 2026-07-28 · Branch: claude/assessment-docs-work-mh32mh · Baseline marker: baseline/pre-arc-797ebad5
Why this file exists. It was meant to be a pull-request body. Neither remote could host one: the GitHub App lacks pull_requests: write (403 "Resource not accessible by integration"), and the agent proxy denies core-gitquarters.acbcontent.org at CONNECT (403, policy). So the index lives in the repository instead — which is the more durable home anyway, since it travels with the commits it describes rather than sitting in a forge that may be re-synced.
Mirror hazard, recorded because it already fired once. On 2026-07-28 the GitHub remote's
mainwas reset backwards from199dc72dto797ebad5, discarding 29 commits of remote history.797ebad5is exactly Forgejo'smain. PerAGENTS.md:280GitHub is a mirror only — so this was almost certainly the mirror being re-synced from the canonical Forgejo remote, which is 29 commits behind. The local clone still had everything and restored GitHub by fast-forward (no force, nothing overwritten). Until Forgejo receives this arc, any mirror re-sync will revert GitHub again.
Verification
uv run bash scripts/ci/local-ci.sh --tier gate @ 199dc72d → EXIT=0, PASSED in 554s
smoke 775 · warmed_session 10 · deductive 520 · teaching 136 — 1,441 tests
The gate went from 2 suites / 721 tests to 4 suites / 1,441 across this arc. Every commit below was gated green before it was pushed.
Commit index
Oldest → newest. Independent = safe to revert alone. Coupled = reverting alone leaves a record contradicting the code.
Wave 0 — record amendments (needed no ruling)
| # | SHA | What it did | Δ | Revert |
|---|---|---|---|---|
| 1 | 9efd5a61 |
PR-1: N-1…N-7 + H-12 applied to the registers | 6f +76/−29 | Independent |
| 2 | 0641e211 |
H-12: the pre-push hook asserts the parity too | 1f +1/−1 | Independent |
| 3 | 0c11b530 |
Merge of PR-1 into the docs branch | — | Merge — revert 1+2 instead |
| 4 | 635f2488 |
Re-verified at main tip; corrected H-8d, which an earlier draft stated wrongly | 3f +4/−4 | Independent |
Track A — ADR-0252 §5 run to a verdict
| # | SHA | What it did | Δ | Revert |
|---|---|---|---|---|
| 5 | 299c92be |
Pre-registers the GO/NO-GO criterion — thresholds as importable constants, no results | 5f +953 | ⚠️ Never revert alone — retroactively unbinds the criterion and voids #6 |
| 6 | 7728a25d |
§5 run → NO-GO. Corpus, experiment, report + pinned digest, verdict doc; two prior GO verdicts on unmerged branches voided (N-8) | 9f +862/−12 | Coupled to 5, 23 |
| 7 | 8f0ccfe3 |
Re-anchored the §5 provenance SHAs to the commits that landed | 3f +13/−4 | Independent |
Triage and two defect fixes
| # | SHA | What it did | Δ | Revert |
|---|---|---|---|---|
| 8 | e025de7a |
External-assessment triage — H-13, H-14, H-8e added; six proposals cleared, each with the refuting file named | 4f +29/−6 | Independent |
| 9 | 71542e61 |
H-13 first attempt — refcount. This fix was WRONG | 4f +177/−9 | ⚠️ Superseded by 10 |
| 10 | 2cfc6ad2 |
H-13 corrected — claimant-set tracking; records why the refcount reasoning failed | 5f +188/−121 | Coupled to 9 |
| 11 | 98a6e8b4 |
PR-9 / H-11 — three silent backstops made visible; byte-identical consumers by construction | 8f +252/−11 | Independent |
Wave 2 — enforcement
| # | SHA | What it did | Δ | Revert |
|---|---|---|---|---|
| 12 | 68e68589 |
PR-4 / G-7 — membership ratchet + full_only_baseline.txt (749). Parity pin withdrawn (N-9) |
8f +1056/−12 | Coupled to 14, 19 |
| 13 | 5f18b7e5 |
PR-6 / G-9 — three doctrine prohibitions enforced by test; four sabotages observed red | 4f +293/−2 | Independent |
| 14 | edf6c2a4 |
The ratchet caught its own author: baseline 749 → 748 | 3f +7/−2 | Coupled to 12 |
| 15 | 39331dbc |
PR-7 / H-7 — M2 trust-boundary table; three boundaries CLOSED, one real delta | 3f +72/−1 | Independent |
| 16 | 3650331b |
Re-stamped the eight cards this arc touched | 10f +47/−1 | Independent |
G-22 — main was red before the arc began
| # | SHA | What it did | Δ | Revert |
|---|---|---|---|---|
| 17 | 913111c0 |
Recorded that main was red and that membership was never the guarantee |
2f +7 | Independent |
| 18 | 0329cc99 |
G-22 fix — two stale ceremony pins; CLAIMS.md had published a superseded evidence digest for two days |
6f +37/−6 | ⚠️ Reverting re-publishes a false digest |
| 19 | 065826c5 |
Pin 3 — test_suite_reachability.py. Measured 21 curated suites, 2 gate-reachable |
4f +201/−2 | Coupled to 12, 28 |
Wave 1 and the docket record
| # | SHA | What it did | Δ | Revert |
|---|---|---|---|---|
| 20 | d0bedfc1 |
R-7 — dead instruments retired; 4 unreferenced aliases deleted (21→17 suites; unreachable 19→15 by deletion) | 8f +74/−24 | Independent |
| 21 | e6a30383 |
Recorded the adopted docket + the 11 standing philosophies in AGENTS.md; fixed 14 PENDING statuses that contradicted reality |
5f +172/−38 | ⚠️ Re-orphans every ruling record |
| 22 | ca5e614e |
G-23 — DOMAIN_PACKS vs manifest domain_id: 7 of 9 bound, 2 absent, P3 passing vacuously. Pin + 4 sabotages |
7f +293/−30 | Independent |
The ruling docket, in its corrected execution order
| # | SHA | Ruling | What it did | Δ | Revert |
|---|---|---|---|---|---|
| 23 | b823d85d |
R-12 | ADR-0146 addendum + ADR-0252 "34 organs" footnote + §5 verdict banner + §6 retirement amendment. Closed H-8a/H-8b/G-15 | 6f +86/−14 | ⚠️ Amends two ratified ADRs |
| 24 | 6c67e88d |
R-3 + R-4 | PR-5 — flag register (32 booleans classified), profiles as units, accrue_realized_knowledge added to the daemon profile. Closed H-8 in full |
13f +498/−16 | ⚠️ Behaviour change — reverting returns the always-on life to consolidating an empty set |
| 25 | 9b190856 |
R-9 + R-2 | PR-11 — the 5,000-beat soak committed as evidence, deterministic_digest pinned, cadence enforced by a source-hash pin rather than a cron that silently does not run |
8f +369/−11 | Independent |
| 26 | 9cf6a164 |
R-13 | PR-12 — the Wilson re-count. 25 licensed shape-bands → 4; the sealer now refuses a padded ledger. wrong stayed 0 |
11f +364/−69 | ⚠️ Serving change — reverting re-grants 21 licences the evidence never supported |
| 27 | 71d6366f |
R-8 | PR-14 — outcome-mix rule as two capabilities. Split, zero bands license — not four. ADR-0264 §5 amended | 7f +261/−8 | ⚠️ Amends a ratified ADR |
| 28 | 294bbe7b |
— | PR-4b — teaching promoted onto the gate (+136 tests); pin 3's own union blind spot closed |
5f +93/−23 | Coupled to 19 |
| 29 | 199dc72d |
R-1/5/6/11 | docs/specs/postures.md — three postures with criteria; R-11's interim gate withdrawn because measurement dissolved its premise |
4f +95/−12 | Independent |
Cherry-pick guidance
Self-contained — take alone: 5f18b7e5 · 39331dbc · 98a6e8b4 · ca5e614e · 9b190856 · 199dc72d
Take as a group or not at all:
299c92be+7728a25d+b823d85d— pre-registration, run, ratification. Any one alone misrepresents the others.71542e61+2cfc6ad2— the wrong fix and its correction.68e68589+edf6c2a4+065826c5+294bbe7b— the ratchet family; each later one repairs a blind spot in an earlier.
Behaviour-changing — read the commit body before reverting: 6c67e88d (daemon profile) · 9cf6a164 (21 licences revoked) · 71d6366f (four licences dissolve)
Rollback targets: 797ebad5 pre-arc · 065826c5 enforcement landed, docket unstarted · d0bedfc1 Wave 1 done, docket unstarted
What the arc produced, in three results
-
ADR-0252 §5 → NO-GO, against a criterion pre-registered before the run. Measured rather than argued: the similarity quotient that delivers attribute-invariance is the same one that annihilates structural contrast. An
add-vs-subtractminimal pair aligns at residual exactly0.0. §6's organ-retirement condition was amended because a NO-GO left it naming a trigger that can never fire. -
Served capability shrank on purpose, twice. Deduction: 25 → 4 licensed bands (replays counted as independent trials; 28 distinct cases inflated to 720). Curriculum: the four predicted licences dissolved to zero — 652–653 correct refusals pooled with 7–8 commitments to clear a 0.000046 margin.
wrongstayed 0 in both. No answer changed; only the claim attached to it. -
R-11's instrumentation dissolved its own question. Nothing is syntactically outside the verified inventory, because
atom_fact's template is{p}— it matches any string. G-2's fabrications are therefore mis-reads of admissible surfaces, not admissions of out-of-inventory constructions, and no gate over the admissibility set can catch them.
Closed: G-5…G-12, G-15, G-17, G-19, G-22, H-1, H-6, H-7, H-8 (all five instances), H-10, H-11, H-13 · Pinned: G-23 · Rulings: 12 executed, 2 stricken, 0 pending.
Not moved, stated plainly: comprehension breadth. The reader is still 19 constructions wide and decides 1.0% of holdout_dev/v1. Everything here is enforcement and evidence machinery. Track B (widening) and Track C (the chooser) are the next arc.
Errors found in my own work, and how
Three, each surfaced by attacking a mechanism rather than trusting it:
- A hollow guard in PR-12 — it compared two functions that agree by construction; the attack it existed to stop walked past it and re-sealed the inflated ledger. Fixed to take the built artifact as input.
- A hollow spot in pin 3 — the two gate scripts could drift apart invisibly (
294bbe7b). - Two invalid measurement passes in R-11, both confident and plausible, caught only by asking whether the matcher could still say no to
"most birds can fly". It could not.
Two earlier statements were also wrong and were corrected in place rather than overwritten: 9efd5a61's H-8d claim (corrected in 635f2488) and 71542e61's H-13 fix (corrected in 2cfc6ad2). Both corrections sit next to the error, because an arc's self-correction chain is its strongest evidence.