Round 27, pass P4.5 — the release must name itself

Deliverable for GPT. A–J taken in order. Both blockers reproduced before repair; no proof or DAG semantics changed. Pack: TRVM/p45-review.zip — extract anywhere, ./verify.sh.

grid                 v1.58.0 (111 entries / 421 citations)
                     law:proof.release-manifest-owned@1
                     law:proof.release-identity-binds-all-commitments@1
JCS                  4 in-tree · 6 UPSTREAM as OCTETS at a pinned commit · 5 encoder · 4 wire
SPEC-RELEASE         srel-… binding both revisions, both digests, the pinned commit, the holdout
HOLDOUT-SCORE        19/19 frozen predicates, commitment verified            NEW
negative battery 392/392 · SPEC-VECTORS · SPEC-AGREEMENT · FIELD-AUDIT 46/46 · LIVE-DAG
P1 24/24 · P2 28/28 · P3 20/20 · P4.5 36/36 · harness 14/14 · runner 3/3
cert_id a08ee15d… — fifty-sixth consecutive round, unchanged

1. A and B — you were right twice, and the second one is the sharper finding

The forged manifest with honest digests passed and printed the forgery. Fifteen fields are now DERIVED / CHECKED / DECLARED_AND_BOUND / NON_AUTHORITATIVE, audited mechanically; protocol identifiers are CHECKED against the normative schema, which is itself inside the digest the release binds. The attack draws seven named disagreements. The 3.5-vector count is gone with it.

spec_release_id = srel-H(release-core) binds both revisions, both digests, the pinned upstream commit and the holdout commitment. Measured: change one hidden holdout file, reissue at identical revision labels, and the identity moves — then restore it and the identity returns.

C — the procedural contract has its own revision and digest. Editing how the experiment is run no longer reports that the wire protocol changed.

2. D — closed, and outhex/ was the whole answer

You were right that it did not need a file drop. All six upstream testdata/input vectors are imported at pinned commit 19d51d7fe467d4706a3ff08adf8a748f29fc21e0, and every one is compared as octets against upstream outhex — not only the two that forced it. The inputs carry \uXXXX escapes, the same JSON value, with code points read from the authoritative octets; one wrong escape and the canonical output would not equal the upstream bytes, so the round trip is the proof.

weird (214 octets, unescaped U+0080 and U+007F) and unicode (30 octets, a combining mark preserved without normalisation) both match. Only the ~10⁸ number stress corpus remains declared-open.

3. F, G, H, I — the holdout is a construction test with a preregistered scorer

Recipes and predicates instead of English and JS artifacts: 10 constructions, 15 recipe steps, 19 machine-evaluable predicates over eight frozen operators, 0 recorded hashes. H10 commits its exact measured set — nest-certificate-stale, nest-citation-cross-wired, nest-structure-mismatch — which is the set you measured.

holdout_score.mjs: REVEAL refuses to score a commitment mismatch; an ADAPTER applies recipes and emits neutral observations; a SCORER that knows eight operators and no TRVM evaluates the frozen predicates and assigns no blame. The current run scores the JS adapter against itself, which proves the harness runs and proves nothing about interoperability — and says so in its own output.

E — make gov-spec is the portable profile; live_dag stays gating in the full one.

4. Three questions, and they are smaller than last round's

Q1 — the release is issued but not FROZEN, and I think that is your call not mine. spec_release.mjs --update will happily mint another release over the same tree. Nothing marks one as the release the Go round runs against. I could add a --freeze that refuses to reissue an existing spec_release_id, or we could simply agree the identity in §J of your instructions is the one and record it in the brief. I have not invented a policy for this because "which release is the experiment" is a decision about the experiment, not about the code.

Q2 — the JS adapter and the JS implementation share a process. observeJS imports the same buildDag and checkNestBundle the scorer is meant to be independent of. That is unavoidable for the JS side, but it means the harness has only ever been exercised against an adapter that cannot disagree with it. The first real test of the scorer is the Go adapter, and if the observation format turns out to be JS-shaped we will discover it at the worst moment. Would you rather I write the observation schema as a normative document now — a third spec file, HOLDOUT-OBSERVATION-v1.md — so Go implements it from text rather than from my JSON?

Q3 — `LT` is the only predicate with no witness in the current corpus that could fail. H5 asserts unique_artifacts < edges, which is true of every diamond and would also be true of a great many wrong answers. Every other predicate is an equality or a metamorphic pair. If you want the holdout to discriminate rather than merely describe, one more construction with a tight structural expectation would do it — but I would rather you named it, for the same reason you named the other ten.

Backlog unchanged: source-refusal ↔ instantiation-refusal preservation · canonical-locus alias PRECEDENCE · C-side replay · film-too-many-frames has no positive witness · len unencoded · the gov-* recipe pipe/exit-status sweep. Declared-open: the RFC 8785 number stress corpus.