DUNIN7 — status brief — v0.12

Version. 0.12 Date. 2026-08-03 Charter. standing-notes/dunin7-standing-authorization-charter-v0_1. Author. Claude Code (verification session). Operator: Marvin Percival. Supersedes. v0.11 at record 25ce17c. All earlier versions stand as siblings; v0.3 remains deliberately absent. Changes from v0.11. CR-2026-162's deferred live pass is run. The item that mattered passes. Two of the three are partially verified, and the boundary is named rather than claimed. One new defect found.

1. The pass, and the fences it ran inside

Walk-audit dev pattern throughout. Throwaway database cr162_livepass, cloned from playground_test and dropped at teardown. Real engine on spare port 8099, a stub on 8098, Operator Layer on 3398 — all torn down, all ports free. playground_dev was never connected to; .env was never modified; the live :8000 process is the same PID it was before. No code, no commits in either repository — both trees clean at 50252e1 and ffc29af.

2. The third thing — the one that mattered — PASSES

Two renders of the same declared type, side by side in the Rendering room:

Render | HTML | #1 of its type | View ↗ | Download | Care guide | Produced Jun 9, 2026 · produced by Goosey
Render | HTML | #2 of its type | View ↗ | Download | Care guide | Produced Jul 21, 2026 · produced by Goosey

Neither is presented as current, and this was audited rather than eyeballed:

• No current-marker in the text. A sweep of the room's rendered text for current, latest, newest, most recent, superseded, out of date, up to date, stale, active, previous, old version returns zero hits.
• No marker in the markup. Both cards carry byte-identical class lists and identical attributes — no state class, no highlight, no data- flag distinguishing them.
• No badge implying rank. Each card's badges are exactly Render, HTML, and its sequence label. Nothing else.
• The sequence number is scoped on screen. It renders as "#1 of its type" and "#2 of its type", not bare #1 / #2 — so it cannot read as a rank across the room.

What the Operator sees is two dated outputs in their type's sequence, and a judgement left to them. That is D-3 as built.

3. The first thing — partially verified, boundary named

Verified live: the engine advertises text/markdown → label "Discovery document (Markdown)" → mode discovery; a .md file uploads and stores as content_type: text/markdown; and the contribution routes to the discovery skill (proven by its own traceback reaching extract_discovery_to_seed_skill).

Verified live at the defect's exact boundary. The defect was response validation, so both changed declarations were exercised over real HTTP against the real engine:

• POST /assertions carrying source_mode: "discovery" → 201 — AddAssertionRequest accepts it.
• GET /assertions/{id} → 200, returning "source_mode":"discovery" — AssertionResponse accepts it. This is the declaration that used to fail and roll the write back.

Not verified: the full skill round trip. extract_discovery_to_seed_skill requires an Anthropic API key and makes an external LLM call. Configuring one is spending (F-5) and was not authorized for a verification, so the contribution could not complete through the skill itself. The schema boundary that was broken is proven fixed; the LLM extraction upstream of it is unexercised.

4. The second thing — not verified, boundary named

A download through the engine's route needs a produced render, and the throwaway holds none (render_events_view count 0). Producing one requires the composition pipeline and an LLM-backed render specialist — the same external-call boundary as §3. The download was not exercised end to end.

What is verified: the route exists with its contract read at Step 0, the adapter targets it, and nine component tests cover the control's behaviour. What is missing is one real byte stream, and it stays owed.

5. New defect found

DiscoveryToSeedNotConfiguredError returns 500, where its siblings return 503. _contribute_file catches TranscriptionNotConfiguredError and ImageDescriptionNotConfiguredError and maps both to 503 Service Unavailable. The discovery equivalent is not in that list, so a missing LLM key on a Markdown contribution surfaces as an unhandled 500 — an internal error where the engine means "not configured".

Found only because the live pass ran. No test covers it: the tests never exercise the unconfigured path. Small, real, and its own build-list item — B-35.

6. Queued

1. B-35 — the 500-versus-503 above.
2. The download round trip and the discovery skill round trip, whenever an engagement with a produced render and a configured key is available. Both are one command each at that point.
3. B-6, the Shaping room — split out at D-1, needs its own Step 0.
4. Q-13 — whether renders need a current-marker, now answerable against a shipped recency display.
5. B-33 / B-34 — neither suite reports a clean run.

7. Acceptance

CR-2026-162's gate is met except for the two round trips named in §3 and §4, both blocked by the same external-call fence rather than by anything in the change. Gate item 7 — the room states nothing the data cannot support — is verified directly and is the item this pass existed to check.

F-1 unchanged: five merges await one restart.

DUNIN7 — Done In Seven LLC — Miami, Florida DUNIN7 — status brief — v0.12 — 2026-08-03 Two outputs, two dates, two sequence numbers, and no verdict between them.