DUNIN7 — status brief — v0.36

Version. 0.36 Date. 2026-08-06 Charter. standing-notes/dunin7-standing-authorization-charter-v0_1, ratified 2026-07-30. Author. Claude Code (inspection session). Operator: Marvin Percival. Supersedes. v0.35 at record be58462. All earlier versions stand as siblings. Changes from v0.35. One inspection landed: the small-work sweep's Step 0, read-only, against thirteen build-list items (B-32, B-35, B-36, B-37, B-38, B-44, B-47, B-52, B-53, B-54, B-55, B-56, B-57). No fix, no change request — the inspection brief's fences barred both.

1. What landed

Record (loomworks-record)

| Commit | Subject | |---|---| | this filing | small-work sweep Step 0 findings v0.1, filed; status brief v0.36 |

Engine / surface

Nothing — this was a read-only inspection. Both repositories confirmed clean and unmodified throughout: engine main at 1706248b626255c09673e9165e8e82030bb7b767, surface main at 1245f1c7584db19921f430b429f69a844221b4ed.

2. Small-work sweep, Step 0 — inspection complete

inspection-briefs/loomworks-small-work-sweep-step-0-inspection-brief-v0_1.md asked, for each of thirteen build-list items, whether it is actually small — one or two files, no design decision, no new mechanism — before any of them are absorbed into a change request. Findings filed at inspection-briefs/loomworks-small-work-sweep-step-0-findings-v0_1.md; copied to ~/Downloads.

Seven are small and ready for a change request:

• Engine CR candidate — B-53 (render list query never selects display_number), B-54 (nothing enforces the number exists — one migration, gated on an unread production-safety check), B-35 (NoCredentialError's exception handler returns 500 where every sibling "no key reachable" path returns 503), B-44 (Shape title routes never call the title-derivation cascade that renders already use — one file, two call sites).
• Surface CR candidate — B-57 (Manifestation room's two-version-numbers sentence — pure string composition), B-38 (RenderingRoom hand-rolls the state branch three sibling rooms get from a shared hook — one file), B-36 (retire-a-render control — the engine route is already complete and correctly irreversible; the surface just never built the button + confirmation dialog).

Six are not small and leave the sweep rather than enlarging it:

• B-37 (failed shaping production is invisible) — spans both repos; the job→shape link column exists but is never populated at creation, and exposing it needs a design choice between a new route and a new response field.
• B-52 (answer sources vanish on reload) — spans both repos; sources were never persisted into conversation history at all, not merely dropped on refresh — needs a new structured_data variant engine-side.
• B-55 ("of its type" names no type) — spans both repos; the wire carries a type id, never a name, on both renders and shapes.
• B-56 (dev sign-in reachable in production) — surface-only but not small; /auth is hardcoded as the real redirect target in 8+ files (session-expiry, sign-out, passkey/recovery-code flows), so hiding it means retargeting all of them, a design decision.
• B-47 (credential issuance screen) — surface-only but not small; the form is two fields, but the credential is a bearer-style claim link granting unmoderated write access, shown once — needs a one-time-reveal pattern and a paired revoke affordance.
• B-32 (re-derive a stale Manifestation) — surface-only but not small; re-deriving means calling a metered, system-key-gated preview step first, which needs a spend-consent surface in front of it.

A correction to the brief's own framing. §6 named B-37 and B-44 as the two items that might span both repositories. B-37 does. B-44 does not — it turned out to be a one-file engine fix. Two items the brief did not flag turned out to span both repos anyway: B-52 and B-55.

One unanticipated finding, not in the build list. The Shape list query has the identical defect as B-53 (render_events_view_query.py's pattern, mirrored in shape_events_view_query.py:109-133 — the SELECT never names display_number there either). Named in the findings document rather than folded into B-53's scope; not filed as its own build-list item here, per the inspection brief's fence against absorbing unscoped work.

One item stays open for the Operator. B-54's fix is code-small, but whether a NOT NULL migration is safe against the live data was not checked — the charter bars an inspection session from touching playground_dev/production beyond read-only metadata. A one-query check, not a build item.

No model call was made directly by the inspecting session; the investigation was parallelized across Claude subagents (read-only, grep/read/bash tool access only) per item cluster, and every subagent claim that made it into the findings document was independently re-verified against the source by the coordinating session before inclusion. Two subagent claims did not survive that check and are corrected in the findings document itself: one incorrectly called a title-derivation function dead code (it is live, called at shape_event_title.py:195), and one sourced an engine field name from the surface's wire-mapping alone rather than the engine's own module (corrected to memory_status.py, no change to the finding's sizing).

3. What is running

Nothing. Record main pushed and origin-verified through this filing. No session in flight.

DUNIN7 — Done In Seven LLC — Miami, Florida DUNIN7 — status brief — v0.36 — 2026-08-06