DUNIN7 — status brief — v0.40

Version. 0.40 Date. 2026-08-07 Charter. standing-notes/dunin7-standing-authorization-charter-v0_1, ratified 2026-07-30. Author. Claude Code (execution session). Operator: Marvin Percival. Supersedes. v0.39 at record 0fc3edd. All earlier versions stand as siblings. Changes from v0.39. The eight-change-request run 175–182. Five landed, two halted on their own findings, one closed as a correctness fix rather than the item it was filed under. Two live defects were fixed. Three items closed by being measured rather than built. Four standing notes and eight record documents filed. Build list moved v0.29 → v0.32.

1. What landed

Engine (loomworks-engine) — 366102d

| Tag | Subject | |---|---| | cr-2026-175-shape-list-display-number | B-59 — shape list never returned its own sequence number | | cr-2026-176-b54-null-display-number-code-fix | B-54 code half — three production paths writing NULL | | cr-2026-177-manifestation-preview-metering | live defect — Manifestation preview was unmetered | | cr-2026-178-b30-regression-trap | B-30 — the W-2 trap made visible to the suite | | cr-2026-179-b12-recall-window | B-12 closes as satisfied; a false comment corrected | | cr-2026-181-b14-spreadsheets-and-decks | B-14 — .xlsx and .pptx ingest |

Surface (loomworks) — 852124b

| Tag | Subject | |---|---| | cr-2026-182-b17-b43 | the workspace selector no longer renders where it does nothing |

Record (loomworks-record) — ace344b

Build list v0.30, v0.31, v0.32. Four completion notes (179, 178, 181, 182). Two Step-1 findings documents (180, 182 within its note). Three scoping/standing notes. Every document mirrored to ~/Downloads.

All three repositories clean and origin-verified. Every tag push-verified at the time it was cut.

2. Two live defects, both found by looking rather than by failure

The Manifestation room was not charging for its model calls. preview_manifestation accepted a spend context and never forwarded it, so every Manifestation LLM call took the unmetered path and the route's own 402 handler could not fire. Found while investigating something else (B-32's consent-surface question), fixed in its own change request ahead of the item that depended on it.

> It survived because the credit gate is unit-tested directly and nothing tested the path to it. A green gate test says the gate works; it never says anything reaches it. The proof that the fix is real is that reverting it returns 502, not 402 — the ungated call ran on to the model and failed somewhere unrelated.

Every Operator-approved reconciliation correction was writing unnumbered records. Three production paths built shapes and renders inline and never called the assignment helper — two in the live reconciliation path registered at application start, one covering every non-produce composition step. B-54's fixture debt had been treated as the whole story since CR-2026-171 reverted its constraint; it was half.

3. Three items closed by being measured, not built

B-12 — the recall window. The measurement that was supposed to justify raising a constant instead corrected the item. Every figure in the prior claim was right: twenty-three engagements hold notes, two hundred thirty-eight between them, the largest holds fifty-eight. They count notes in every state, and recall reads only the settled ones. That engagement holds ten settled; the largest anyone works in holds thirteen, against a window of fifty. The window has never bound anything. No constant was raised.

> Why the error was invisible: the counts were right. Only the quantity was wrong.

B-30 — the trap behind B-27. Nothing to repair: the dependency is correct for routes that carry the path segment, and a walk of all 228 routes found no live instance. The defect the item names is the suite's blindness, and that is what was fixed — closed as detectability rather than repair, on explicit direction, and recorded as such so the original wording is not later read as satisfied.

B-13 — ask your engagement. Also mostly built: the question, the truthful answer, the stated boundary and the sources are all in place. One clause of its own wording fails — "every source shown" — because the standing context block enters every prompt and never appears in sources. B-13 stays open, now blocked on a precisely-named thing rather than on B-12.

4. What the record gained

Four standing notes, each from a defect rather than from theory:

1. Build the spend context at the gate, not at the route. Shaping and Rendering are safe because their shape gives no opportunity to be wrong; Manifestation threaded a value through two layers and one dropped it. Prefer designs where the unsafe state is unconstructible over designs where it is merely unwritten.
2. The mechanism that makes something safe in one dimension hides the failure in another. Test isolation blinded the suite to uncallable routes; unit coverage of the credit gate said nothing about whether anything reached it. Both fail open, and both were produced by machinery working exactly as designed.
3. A check is not trusted until it has been observed failing for the reason it exists. Four instances in this run of a check reporting green while unable to catch anything. Break-and-observe caught three; the fourth — a baseline drifted looser than reality — was caught by reading a diff, which is why the note says break-and-observe is necessary and not sufficient.
4. Regenerate the mypy floor only with the diff read. A ratchet that can loosen without anyone noticing is not a ratchet. Filed in the mechanism's own documentation and cross-referenced from the failing CI step.

Six new build-list items that previously existed only as comments, notes, or nobody's job — B-59 through B-64. The one that prompted the sweep, B-60, lived as an in-code comment describing itself as a follow-on. A comment is not tracking.

5. Two change requests halted, and why that is the system working

CR-2026-180 (B-13) halted at its Step 1 read rather than building what was already there. It also carried an acceptance criterion — non-suppressible AI origin marking — that was written from recollection rather than from the seed. The seed is silent on it. Recorded plainly, because a change request asserting a commitment is exactly what a later session would cite as evidence the commitment exists.

CR-2026-182 (B-17) halted on the same shape. The portfolio filter was built, then unbuilt: one redesign deleted the section that consumed it and left the engine filter, the adapter, the provider and the chooser standing. The chooser then offered a choice that changed nothing, on every screen it appeared on, for a full redesign cycle. What remains is engine work across three cross-project routes, not the "one parameter" the list had claimed for twenty versions.

> The lesson recorded with it. Keeping a control because we intend to make it true later is the reasoning that produced the defect. The control is gone; the component is kept, so restoring it is one line once there is something to filter.

6. Decision queue

Two items want you, neither urgently.

B-63 — nothing marks an answer as written by a model. No marking exists in either repository and the seed does not say whether one should. Seed-level; blocks nothing.

B-61 — an answer can be shaped by material it cannot cite. This one blocks B-13, and its live-vs-latent question is already settled: a recall turn is the case, by construction. Not a sizing question — making the two windows match, in either direction, leaves it exactly as it is.

Everything else proceeds without you. B-17 and B-43 are both returned to the list with what is known about them; B-43's costing was re-checked and still holds.

7. What is running

Nothing. Engine 366102d, surface 852124b, record ace344b — all clean, all origin-verified. No session in flight.

DUNIN7 — Done In Seven LLC — Miami, Florida DUNIN7 — status brief — v0.40 — 2026-08-07