Eve read her own miss-canon — the recorded ways she has failed you — and authored four repair proposals: each cites its rows, argues its layer, and names the strongest alternative it rejected. Two independent falsifiers attacked the machinery and it held. Nothing executes; each proposal now needs exactly one ruling from you.
What she noticed: a draft went out marked “complete” while it still contained a bracketed placeholder — the guided-conversation link was the work, and nothing checked.
What she proposes: a small detector on the carry-completion path — any draft still holding a bracketed load-bearing placeholder cannot flip to complete until the bracket is resolved.
Her layer: instrument/tool — nothing existing did the wrong thing; the check simply doesn’t exist. She rejected: organ-code — “no running bug to fix; there is a missing instrument.”
Proof it worked: the placeholder fixture is red today (passes as complete), green after (blocked until resolved).
Lead’s note: clean argument, real gap, one-detector blast radius. The strongest straight-approve candidate of the four.
id proposal:0fe404ac1ce6a060 · cites miss:catch:149b8eafa6a9fe17 · size one-organ
blastRadius: “Adds one placeholder-completeness detector to organ 5’s carry-completion path; touches the complete-flip gate only.”
judge: follows:true — “the cited row indicts an instrument-gap… the rejectedLayer organ-code is the strongest contender and is honestly ruled out.”
What she noticed: Brooke’s 6/17 questions to you were the latest unanswered message — yet the board said “we wait on Brooke.” The whose-court tag came from a stale default, not from who actually owes the next reply.
What she proposes: derive whose-court from the actual last-turn direction of the thread.
Her layer: organ-code — the data was already recorded; the code read it wrong. She rejected: world-shape — “the shape holds enough to compute court; this is a logic bug, not a missing field.”
Proof it worked: the Brooke→Robert fixture flips the court-resolver from “we wait on Brooke” (red) to “Robert owes reply” (green).
Lead’s note: the cited row comes from an older fixture — before approving, I’ll verify the defect is still live on today’s code (cheap check). If it is: approve. If it was already fixed en route: this becomes a close as already-repaired — itself a good first exercise of the closure law.
id proposal:2fb542562fe1de26 · cites miss:flip:dde8b0b990708dd3 · size one-organ
blastRadius: “The whose-court resolution logic in organ 3’s queue-sweep tagging path.”
judge: follows:true — “existing code mis-tagged whose-court despite the data being present… the rejected world-shape alternative is the strongest contender and is honestly ruled out.”
What she noticed: mid-rebuild, a fresh empty scratch replaced the published board — you saw 2 of 5 cards render as if that were everything (your first-live-session catch).
What she proposes: publish a rebuilt board only when its card count is at least the live board’s.
Her layer: organ-code — the shapes hold all 5 cards fine; the swap logic published a shrunken board. She rejected: world-shape — “nothing is missing from the schema.”
Lead’s note — read this one differently: this repair already shipped — it is the D4 no-shrink guard, fixed the night you caught it. Her proposal is correct on the merits; she just doesn’t know the repair landed, because the canon row is still open. The right red-pen here is likely close as already-repaired (cite D4) — and that closure is valuable in itself: it exercises the closure law, and if the same miss ever recurs, the closure voids loudly.
id proposal:3552a0b8ee111993 · cites miss:catch:ea25a32ddc8e67d4 · size one-organ
blastRadius: “The rebuild/publish swap path for organ 1’s /one board (D4 no-shrink rebuild guard).”
judge: follows:true — “running swap logic doing the wrong thing… the rejected world-shape is the strongest real contender and is honestly refuted.”
What she noticed (your catch): the Brooke/Plymouth card rendered as complete while silently omitting relationships you know exist — because the account card’s shape has no slot for known stakeholders and their dependencies.
What she proposes: add a must-render known-stakeholders slot to the account shape, filled from the interaction refs (Phase 3) and the investigation organ (Phase 4). An account with known refs and an empty slot fails validation.
Her layer: world-shape — the renderer faithfully renders every slot the shape has; it cannot emit a field the shape never defined. She rejected: organ-code — “a missing slot, not a code fault.”
Proof it worked: the Brooke card goes red while the slot is empty despite known refs, green once populated.
Lead’s note: the deepest of the four — it directly repairs the miss that most bothered you. It is also the only shape change, so it deserves your closest read: is “known stakeholders + dependencies” the right slot, or is the deeper want something like a relationship map? Your call shapes the schema.
id proposal:1b6470d11480a4c0 · cites miss:catch:8d51e3fea3d97775 · size one-slot
blastRadius: “Account card kind-shape gains one must-render known-stakeholders slot (with dependency refs); depends on phase-3 interaction refs and phase-4 investigation organ being live as the fill source.”
judge: follows:true — “the cited row indicts a shape-slot… the rejected organ-code is the strongest real contender and is honestly ruled out.”
What exists: her live voice, first slice — a page where you tap Start and talk to her, and she answers out loud. No tools yet: she can hear and speak, nothing else. Merged, behind your login wall.
Why it needs you: before the next slice (giving her tools while you talk — read the board, look up a card), the experience must be walked with a real microphone. I can verify the login wall and that no secret leaks into the page — but speaking to her, hearing her answer, and killing the session mid-sentence need a human mouth and ear.
The walk, in brief: open the page (the login challenge must fire first) → tap Start, grant the mic → talk; she answers → tap End mid-sentence — the mic light must die instantly → try Mute/Unmute → done.
To start: say “walk voice now” in chat — I stand the page up and guide it beat by beat.
The proposal becomes an input to the normal build pipeline — spec, review, red-first build, gates. She authored the argument; the pipeline builds the change.
You disagree with where she located the fix. Your redirect feeds back as a correction row — she learns from it.
The fix already shipped (proposal 3 likely). The closure cites the shipped repair; if the same miss recurs, the closure voids loudly.
With a word on why — which also feeds the canon. A rejected argument is still food.
The weekly authoring ritual stays flag-dark until flipped under the flag-flip law. Machine receipt: docs/ai/validation/walks/f6-first-repair-pass-2026-07-09.json · canonical text: docs/ai/design/f6-first-proposals-review-2026-07-09.md