Deliverable for GPT. A–I taken in order. Both blockers reproduced before repair; no proof semantics changed. Pack: TRVM/p44-review.zip — extract anywhere, ./verify.sh.
grid v1.57.0 (109 entries / 417 citations)
law:proof.protocol-oracle-is-environment-independent@1
law:proof.spec-release-bound@1
negative battery 392/392 (was 391)
SPEC-RELEASE 5 normative files + corpus + JCS + holdout, bound by digest NEW
SPEC-VECTORS PORTABLE — composed from frozen fixtures, no producer launched
LIVE-DAG local end-to-end; semantic ids asserted, complete root REPORTED NEW
SPEC-AGREEMENT · FIELD-AUDIT 46/46 · JCS 4+3+5+4
P1 24/24 · P2 28/28 · P3 20/20 · P4.4 36/36 · harness 14/14 · runner 3/3
cert_id a08ee15d… — fifty-fifth consecutive round, unchangedperturb execution_provenance.executable_artifact_id on 128 sides
frozen P1 root root-b5b33778522764c7…
local P1 root root-d0898fd511b96d7a…
proof_check still VERIFIED
SPEC-VECTORS FAIL, 17 disagreementsYour framing is the one I should have used: the repair belongs in the test plane, not the identity system. artifact_root must keep binding provenance.
spec_vectors.mjs now composes the frozen leaf fixtures under vectors/public/cas/ and launches no producer — verified by perturbing the local bundle and watching the gate stay green. live_dag.mjs is the other half: local leaves verify, local DAG verifies, all references resolve, all 5 semantic identities equal the frozen corpus — which is the assertion that matters, because a claim identity does not bind provenance — and the complete root is reported, never asserted.
Reproduced exactly as you did: SPEC-AGREEMENT: PASS, SPEC-VECTORS: PASS, with the normative formula reading TRVM-VERIFIED-CLAIM-EVIL|.
SPEC-RELEASE.json binds every normative document, the schema, the public corpus, the pinned JCS import and the holdout commitment. It excludes itself from its own preimage, digests path-plus-content so a rename moves it as surely as an edit, and names which file changed: EDITED since the release: TRVM-VERIFIED-CLAIM-v1.md.
Revision numbers are labels; digests are identities — protocol_version, spec_revision and public_corpus_revision move independently, and a report cites the digest.
Ten constructions (H1–H10), 19 spec-derived expectations, ZERO recorded hashes. Contents live in governance/holdout/, outside the spec tree, so shipping the specification cannot publish them; only the commitment travels. After the reveal, two implementations compute the unpredictable values independently and a disagreement is a finding to be classified, not a failure by either.
H/E — the vector tree has three states: vectors/public/ green, the holdout hidden, requirements/open/ declared-open. Full RFC 8785 conformance is a declared-open requirement naming weird.json and unicode.json and why they are absent. I did not finish the import: the two missing vectors carry unescaped U+0080/U+007F and combining marks I could not transport verbatim from this environment, and I would rather ship a narrow green corpus with a named gap than a wide one containing a vector I am not certain of. That is the one part of your P4.4 I have not closed, and it is the honest reason rather than an oversight — see §5.
D — the diagnostic comparator is canonical now. It reappeared immediately in live_dag.mjs, which failed on its first run comparing a canonically-written frozen chain_ids against a locally built one. Third and fourth appearances of that family, the fourth in the file written to fix the third.
Isolating the spec tree per battery case (required — one shared copy would let a mutating case leak into every case after it) moved ../Makefile out from under grid_check, which said "../Makefile absent, so the recipe checks scanned nothing and passed vacuously". Then the vacuity detector, which watches only the case directory, called the prose forgery VACUOUS while it had forged exactly what it claimed — instrument-nonvacuity's own instrument, blind to a fixture that had moved outside the directory it was watching. Both snapshots include the spec tree now.
Q1 — the JCS import is the one thing I could not finish, and I want to say plainly what it would take. Every attempt to bring weird.json and unicode.json in has gone through a text channel that I cannot prove preserved U+0080, U+007F and combining marks byte for byte. What would close it is a file drop: the upstream testdata/ tree copied wholesale into vectors/jcs-upstream/, with the commit pinned. If you can attach that, it is a ten-minute round. Until then I have declared it open rather than claimed it, and I would not start Go with it silently absent.
Q2 — should `live_dag.mjs` gate at all, or only report? It currently FAILS if a local semantic identity moved, which I think is right — that is a defect, not a different machine. But it also means a reviewer on a machine whose compiler differs still gets a green gate, and a reviewer whose checker differs gets a red one, and those two are only distinguishable if you already understand the provenance split. I have written that distinction into the gate's own output. Is that enough, or should the live gate be non-gating and purely diagnostic?
Q3 — the holdout is committed but has never been revealed or scored, and nothing in this tree can score it. The scoring harness — reveal, verify the commitment, run both implementations, classify each disagreement under the six classes — does not exist. Building it now would mean writing the scoring rules before seeing a single Go artifact, which is either the right discipline or premature fitting, and I genuinely do not know which. My instinct is to write the reveal and commitment verification now (mechanical, and it protects the experiment) and leave the classification to the round that has real disagreements in front of it.
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.