Version. 0.41
Date. 2026-08-08
Charter. standing-notes/dunin7-standing-authorization-charter-v0_1, ratified 2026-07-30.
Author. Claude Code (execution session). Operator: Marvin Percival.
Supersedes. v0.40 at record 9ba40e5. All earlier versions stand as siblings.
Changes from v0.40. The six-change-request run 183–188. Five landed, one halted at its read. Two seed-level gaps found — one of them a commitment nobody had written down. A build-list sweep found six items marked READY that were already done. Build list moved v0.34 → v0.40.
loomworks-engine) — ad83960
| Tag | Subject |
|---|---|
| cr-2026-183-b61-citability | notes the answer cannot cite no longer reach the model |
| cr-2026-184-b37-failed-production-visible | a failed production stops vanishing from the active list |
| cr-2026-185-b52-persist-answer-sources | an answer's sources survive a reload |
| cr-2026-186-b55-type-names | the wire carries a type name, not only an id |
loomworks) — 0dd7dd9
| Tag | Subject |
|---|---|
| cr-2026-184-… | the failed state renders distinctly; no retry control, because there is no retry |
| cr-2026-185-… | sources rebuilt from history on reload |
| cr-2026-186-… | the type is named, or the phrase does not appear |
| cr-2026-187-b56-dev-signin-unreachable | the dev sign-in page 404s in production |
loomworks-record) — 7b157abBuild list v0.35 → v0.40. Three completion records, one halted-with-findings. One change-request amendment each for 183, 184, 188. Architecture specification v0.7. Standing note v0.2.
All three repositories clean and origin-verified. Every tag push-verified when cut.
The posture the whole marking question rested on does not exist. The foundation document defers an "AI-as-faithful-clerk posture" and treats it as settled — the rule that the Operator approves state transitions is described as consonant with it. It appears in that one sentence and nowhere else. The Operator confirms he does not recall writing it and considers it a good framing, which is not the same as having decided it.
> It had nonetheless been asserted as settled, in a change request's acceptance criteria and in the Operator's own working notes. That assertion was unfounded and is recorded as a correction. The consequence is not cosmetic: marking model-written answers is now a new commitment if adopted, not a forgotten one. A forgotten commitment is recovered; a new one is decided. Treating the second as the first would have skipped the deciding.
And the deferral it belongs to has no destination. The foundation document sends four topics — the clerk posture, the six-stage lifecycle, the badge mechanism, retirement-with-dignity — to "the philosophy and architecture document." The architecture specification contains none of them, and no philosophy document exists in the record.
Four empty sections now exist for them (architecture specification v0.7), each carrying what the foundation document promises it explains and the questions that explanation must answer. No content was written — none of it is inferable, and the lifecycle in particular was deliberately not read off the running system, which would have recorded an implementation detail as an intention.
The foundation document still points at an empty destination, deliberately. Correcting the pointer now would move it rather than remove it. That correction is its own act, once the sections hold something.
A sweep of the build list — prompted by two stale pointers found by accident — found six items marked READY that were already done, closed by change requests that predate six versions of the list.
> Why incidental discovery kept missing them. Nobody follows a status. A pointer gets walked: reading one item sends you to another, and a stale pointer is caught on arrival. A status is read only when somebody opens the item. The cost was already paid five times this run — B-12, B-13, B-17, B-43 and B-37's engine half were each opened for building and each turned out already built or already satisfied.
And the list contradicted itself. Closures had been recorded in a grouped summary line and not in the items' own entries, so two places disagreed and either read as authoritative alone. That contradiction was introduced while fixing a different staleness — the same shape as the walk-audit finding already recorded in the list: the sweep reported the files containing stale text and never asked whether those files were current.
B-19 now carries a standing check: when an item closes, update every place it appears, and check a status against the tag that closed it rather than against the previous version of the list.
Three of the six were written against a plausible mental model that the code did not support, and the read-first step caught each one:
Two steps were withdrawn on finding rather than skipped, and are recorded that way so nobody reads them as shortfall.
The standing note on observing failure was split into two disciplines after this run:
The second was earned. While proving a new test could fail, a restore put back only one of two widened filters, leaving a count query narrow. Break-and-observe had already done its job by then. What caught the half-fix was what the assertion said — that the count must agree with the list. A bare equality assert would have failed identically and communicated nothing.
Guards that were not mine also earned their keep: an adapter's deep-equal contract test flagged a new field; the vocabulary wall caught a comment that used the substrate's shorthand to explain that the Operator never sees it; the non-erasure guard matched a variable named for truncation, and was renamed around rather than exempted.
Two items want you. Neither blocks anything being built.
B-65 — the four scaffolded sections await your content. Only you can say whether the clerk posture exists outside the record or was never written. Until they hold something, the foundation document's pointer stays deliberately uncorrected.
B-63 — marking model-written answers. Correctly classified now: a new commitment if adopted. What it would cost is real — a mark on every answer a model writes, non-suppressible, permanently.
B-69 is newly filed and carries a standing instruction for every change request until it resolves: contact points where something assumes a seed is fixed get recorded under a seed-mutability impact heading. Do not design for it, do not build around it — the point is an inventory rather than a search, so that when it is scoped the question is answerable from a list written as the code was touched.
Nothing. Engine ad83960, surface 0dd7dd9, record 7b157ab — all clean, all origin-verified. Build list at v0.40. No session in flight.
DUNIN7 — Done In Seven LLC — Miami, Florida DUNIN7 — status brief — v0.41 — 2026-08-08