DUNIN7 — status brief — v0.46

Version. 0.46 Date. 2026-08-26 Charter. standing-notes/dunin7-standing-authorization-charter-v0_1, ratified 2026-07-30. Author. Claude Code (execution session, completion-program Wave 1). Operator: Marvin Percival. Supersedes. v0.45. All earlier versions stand as siblings. Changes from v0.45. Wave 1 of the completion program executed in one session under the ruled decision sheet (P-1 yes, W-1/W-2 as ruled, RE-1 untouched). Six CRs shipped, three items found already done, two parked to the queue as the decisions they are. Build list moved v1.07 → v1.08. One deploy and one walk are owed and named in §4.

1. What landed — the finish-line thread

Six change requests, drafted in-session under P-1, each against a settled finding, each merged on a green suite (3,988 → 4,015 tests over the day, zero failures at every merge).

| CR | Item | Merge | What changed | |---|---|---|---| | CR-2026-236 | B-79 | engine a0da312 | The shape-confirmed dispatch hook is registered at startup. Confirm now enqueues its declared renders — the machinery existed since CR-2026-021 and had never once run in production, because only tests called the registrar. The new test pins the property that failed: the call site's existence. | | CR-2026-237 | B-78 | engine 56ccbba | The Manifestation write through conversation: derive_manifestation capability, request_manifestation intent, the CR-2026-189 stated-facts consent card (change counts free, price explicitly unknowable), a per_action_only dispatcher that runs the metered preview and stores exactly what it produced — the room's own one-press adjacency, transferred. | | CR-2026-238 | B-37 | engine ef5c4f6 + surface a172a43 | Failed productions have a place to appear. Step 0 corrected the filing: no held ShapeEvent is ever written — a failure left nothing, and its stored reason had no list surface. Now the engine lists an engagement's shaping jobs and the room names running and failed productions, over an empty room too. | | CR-2026-239 | B-97 | engine 482631c | The six quoting acks cut on word boundaries — CR-2026-209's discipline borrowed, as the filing instructed. | | CR-2026-240 | B-100 | surface c0de8a5 | The findings answer box moved into the working surface — the conversation column the composer trained the Operator's hands on. Placement only, per the filing's own ruling. | | CR-2026-241 | B-108 | engine b77339b + 72e4499 | The save-offer "yes" routes to commit_assertion deterministically ahead of the classifier; the handler is untouched (the recorded constraint, honored by construction). The delegation_held grant ack joined the personal redirect; the two assent lexicons became one. |

Behavior change the deploy carries, named because it is the point: from the first restart with CR-2026-236, confirming a shape on an engagement with an active declared render-type and a registered specialist auto-enqueues render production and spends against that engagement's key (trigger="declared_auto_on_shape_confirmed"). That is W-1's ruled meaning of confirm, not a side effect.

2. The already-done items — the report the Operator asked for

Wave 3 had two; Wave 1 had three. The pattern repeats: entries filed before a fix landed elsewhere never got flipped.

B-13 (ask your engagement) — verified live on the serving perimeter this morning, riding CR-2026-225's first restart: a real /operator/converse recall question returned the semantic path, ranked content, stated scope, populated sources. Clause 5, whole, in production.
B-52 (sources vanish on reload) — CR-2026-185 shipped both halves on 2026-08-13 (engine 1c3d228, surface 73789f4). The build-list rollup still said READY.
B-94 (commit sentence swallows the name) — died with feat-create-stage-polish Decision B (354429f, merged 39cee7f, 2026-08-25): the sentence is constant now, exactly the entry's third proposed shape.

3. What a Step 0 found that changes a later wave's shape

B-114 is Wave-1 material and is not small. The W-5 free answer came back positive: the Companion's produce_specification path requires a grammar-carrying shape-type and produces {"text": …} — so every conversational spec confirm on the walk path demands an override rationale. The fix is an executor output-contract design (keyed output vs a text-aware evaluator vs per-grammar evaluators). A scoping note is requested (§5). Whoever plans Wave 4's walk script should assume the override-rationale step exists until this ships.
B-37's filing was wrong about the mechanism in a way that matters later: there is no stuck held entry — no ShapeEvent exists at all until success. Anything downstream that assumes "requested work is visible somewhere" (the walk script, dashboards) should lean on the new shaping-jobs list, not the shape-events list.
The shape_produced and drift post-append hooks remain deliberately unregistered. CR-2026-236 registered only the confirmed hook; the siblings would switch on auto-opened confirmation considerations and per-append LLM drift dispatch — both unruled, the drift one a spend change. If a later wave wants them, that is a decision, not a registration.
A process finding from CR-2026-241: the mypy ratchet's failure was masked because the pre-push sweep piped the gate through tail, which replaced its exit code. Caught in-session on the post-merge re-run, fixed at 72e4499 within minutes. The rule: a gate's exit code must never be read through a pipe.

4. Owed — the Operator's acts

One deploy, batched per P-2: restart the engine service onto main at 72e4499 (launchctl kickstart -k gui/$(id -u)/com.dunin7.loomworks-engine), and rebuild the Operator Layer onto main at c0de8a5 (which also carries the still-undeployed 39cee7f create-stage work and CR-2026-238/240's surface halves). After the engine restart, a confirm on a specialist-backed engagement will enqueue renders — see §1's behavior note.
The Wave-1 walk: seed to render on a fresh engagement, every clause of the finish line once. Expect the B-114 override-rationale step at spec confirm (§3.1). The create-stage acceptance walks (queued-directions/loomworks-create-stage-browser-acceptance-work-item-v0_1) can ride the same sitting.

5. Decision queue

RE-1 (the roster exception) stands open, untouched — nothing in Wave 1 depended on it. Two entries added:

B-103 — what should an identical duplicate capture do against a standing fact? Step 1 is answered (the matcher reads the record, not the conversation; identical duplicates pass because its semantics are "changed value" by design — a correct null, not a scope bug). Options: offer-to-skip ("this exactly matches #N — hold it anyway?"), hold-with-linkage, refuse. Recommended default: offer-to-skip via a deterministic exact-match check (normalized equality, no LLM), composed server-side — the walk's praised supersede-offer pattern extended to identity. Blocks nothing; the wave completed without it.
B-114 — scoping request (P-1's unscoped-item rule): the executor output-contract design above. Blocks the walk's spec-confirm step being rationale-free; blocks nothing else.

Standing entries (B-65, B-63, B-69) carry forward unchanged.

6. What is running

Engine serving process: pid 99811 (started 2026-08-26 12:30), carrying through CR-2026-235 — eight merges behind main as of this brief (236 through the 241 follow-up). The deploy in §4 closes the gap.

DUNIN7 — Done In Seven LLC — Miami, Florida DUNIN7 — status brief — v0.46 — 2026-08-26