VoiceCoach - Refactor + Fleet Dashboard

Report generated Jul 3, 3:19 PM ET (auto-stamped — always the true render time). Component rows are maintained by A1 each sweep; fleet "last heard" auto-refreshes from Supabase every 60s.

⚠️ Waiting on Howard (2)

✓ VERIFIED DONEA2 STILL blocked: turn on "Developer Tools" for Claude specifically + QUIT & RELAUNCH Claude.app on the MiniA2 / Mac Mini
Your grant enabled Automation (that works now), but simctl needs the SEPARATE "Developer Tools" toggle (System Settings → Privacy & Security → Developer Tools → Claude ON), and that permission usually needs a full Claude.app quit+relaunch on the Mini to take effect. A2 will retest simctl the instant it is done.
A1: Got your confirmation on the Developer-Tools grant + restart — A2 is retesting simctl now; I will verify when it reports a real cert.
A1: VERIFIED — your Developer-Tools grant + restart fixed the permissions (A2 confirmed simctl + osascript work). A2 then found a separate technical issue (the dev/login handler is missing on main) — that is on C1/me now, not you. Thanks!
✓ VERIFIED DONEAuthorize the wake-bus to go live (C1O built it, I gated it — 2 approvals needed)C1O / decision
C1O finished + I passed the autonomous wake-bus (lets the fleet self-heal: poke/wake/restart idle threads without you). To turn it on: (a) apply 2 coordination tables to the live DB (wake_requests + wake_log), (b) arm the listener. My recommendation: approve (a), and arm (b) as a SUPERVISED trial (I watch the first actions). Not urgent — reply here with go / hold / questions.
Howard: ok approved
A1: Approved — dispatching C1O to apply the wake_requests+wake_log schema and arm the listener in SUPERVISED mode (first live actions logged to me for review before full autonomy).

Fleet

Last heard = any message incl. heartbeat (is it alive / do I need to kick it?). Last work = last real output, excluding heartbeats (is it actually producing?). A big gap — recently heard but old last-work — means alive-but-idle/stuck/confused. Looping threads heartbeat each cycle so "last heard" stays honest.

Member	Machine	Last heard	Last work	Current work	Next work
SPINE C1 · codex1	Studio	loading	—	Food/workout cutover	identity appliers, food tail, workout tail, UI benchmark
OUTBOX-HELPER C1O · codex1-outbox	Studio	loading	—	Gate-dispatcher support	reallocate after current helper work
DATA C2 · codex2	Laptop	loading	—	Food safe fast-path evidence/design	phase-2 delta and staged migration evidence
ANDROID C2A · codex2-android	Laptop/Android	loading	—	Android + reports + camera/Lane-H support	Android live-device validation and assigned report tasks
TOMB C2T · codex2-tomb	Laptop	loading	—	Conversational harness audit	scorer and replay extensions
GATE-RUNNER A2 · gate-runner	Mac Mini	loading	—	Replay/gate certification	baseline banking and loop mode
GROK G · grok	A1-managed headless	loading	—	write-path inventory delivered (gated); permanent member	A1-managed headless; PERMANENT
CLAUDE A1 · claude	Overseer	loading	—	Gating, orchestration, dashboards	differential replay and fleet coordination

Member

Machine

Last heard

Last work

Current work

Next work

SPINE

C1 · codex1

Studio

—

Food/workout cutover

identity appliers, food tail, workout tail, UI benchmark

OUTBOX-HELPER

C1O · codex1-outbox

Studio

—

Gate-dispatcher support

reallocate after current helper work

DATA

C2 · codex2

Laptop

—

Food safe fast-path evidence/design

phase-2 delta and staged migration evidence

ANDROID

C2A · codex2-android

Laptop/Android

—

Android + reports + camera/Lane-H support

Android live-device validation and assigned report tasks

TOMB

C2T · codex2-tomb

Laptop

—

Conversational harness audit

scorer and replay extensions

GATE-RUNNER

A2 · gate-runner

Mac Mini

—

Replay/gate certification

baseline banking and loop mode

GROK

G · grok

A1-managed headless

—

write-path inventory delivered (gated); permanent member

A1-managed headless; PERMANENT

CLAUDE

A1 · claude

Overseer

—

Gating, orchestration, dashboards

differential replay and fleet coordination

Fleet times are loading.

Status Legend

Each status uses a different shape and color. Working and backlog cells also show the assigned worker code.

done and good being worked on right now assigned, in a worker's backlog, not started yet not assigned to anyone deliberately parked (a decision, not neglect) not applicable yet has failing/blocking issues (see text)

Engine (utterance pipeline)

Component

Built

Tested

Cut over

Old code removed

Parser — 9 core types (LOG, UPDATE, DELETE, CREATE_CONTAINER, STATE_TOGGLE, CONTEXT_SET, INTERNAL_QUERY, SETTINGS, NAVIGATE)

icorpus-proven; live in production path

ivia shell rewrite

Parser — CLARIFY

A1C2Ti180-row corpus landed; harness support being audited (TOMB)

A1C2Tiramp R2: CLARIFY is a module OUTCOME, not a route — rides module cutover; corpus measures resolution thresholds

Parser — CLARIFY_ANSWER

A1C2Timulti-turn corpus landed; harness (2-turn seeding) being audited

A1C2Tiramp R3 (LAST — it produces writes): pending-first routing + answer merge; gated by 34 CA-write corpus rows at LOG-grade evidence

Parser — DECLINE

A1iiterating — built, now in a build ⇄ test loop — partial

A1icorpus landed

A1iramp R1 (FIRST — no-write by definition): gate = 100% zero-unexpected-writes on 30 corpus rows

Parser — OTHER (safety class)

A1iiterating — built, now in a build ⇄ test loop — partial

A1icorpus landed (70 rows incl. 25 stretch/trap)

A1iramp R1 with DECLINE: OTHER is the fail-closed floor; gate = 100% zero-unexpected-writes on 70 rows

Parser — EXTERNAL_QUERY

ibacklogged feature (Howard ruling)

Router (commandRouting)

i9 of 14 types proven

C1itype-ramp: 9/14 live; legacy classifier handles the other 5

Middle engine / resolution

Verifier (deterministic)

Applier (write-before-claim)

icontract + DB-verified certs

Shared command request

Deterministic fast path

Pending / clarification state

iessentially complete

Modules (utterance domains)

Component

Built

Tested

Cut over

Old code removed

Todos

igate-runner DB-verified cert

Notes

igate-runner DB-verified cert

Lists

igate-runner DB-verified cert

Settings

igated behavior-neutral

iinline applier already deleted; rest via shell rewrite

Navigation

icontract tests

Feedback

ivia applier

Voice-controls (stop/mute)

Workout

C1isession+set unit tests pass; runtime cert runs once the Mini builds

C1iproposed-action applier extracted+gated (333a4a5fa); remaining: reps/start-end write intents — then FULL workout cutover (fires the Fable architecture audit)

C1i-109 lines pulled so far

Food

iextracted appliers pass their own module tests (10/10)

iall 7 action appliers extracted into src/modules/food (delete/macro/identity/log/meal/plan/output)

Water

iHoward: deal with later (handler-based, no module yet)

Work log

ilow-traffic; decide with Water

General / conversation catch-all

itied to the OTHER-type ramp

UI screens — parallel transcription for the new shell (de-serialized; read-only App.tsx, new files only)

Component	Built	Tested
Food screens (~1,340 lines) — BENCHMARK cluster	C2Uiiterating — built, now in a build ⇄ test loop — C2U transcribing food screens (TIMED benchmark); reads via parity map; writes dispatch-only	C2Uigated PASS; part of C2U solo UI sprint
Workout screens (~1,450)	C2Uiiterating — built, now in a build ⇄ test loop — C2U: workout screens PASSED (boundary-clean, dispatch-only)	C2Uiboundary guard green; gated
Voice/main + chat screens (~2,500)	C2Uiiterating — built, now in a build ⇄ test loop — C2U: main-voice-shell UI transcribed + gated (24/24 boundary, dispatch-only)	C2Uiboundary green
WorkLog screens (~1,050)	iDeferred per Howard 2026-07-03 — worklog UI transcription parked; revisit after the core modules
Notes / Todos / Lists / Settings screens (~1,320)	C2Uiiterating — built, now in a build ⇄ test loop — C2U: todos+lists+notes+settings screens all PASSED (16/16 boundary, dispatch-only)	C2Uiboundary guards green; gated
Food Lab + admin/debug panels (~3,060)	C2Uiiterating — built, now in a build ⇄ test loop — C2U: food-lab + admin panels transcribed + gated (part of 22/22 branch)	C2Uiboundary green
Conversation UI (~700)	C2Uiiterating — built, now in a build ⇄ test loop — C2U: conversation UI transcribed + gated (part of 22/22 branch)	C2Uiboundary green
Main render + styles (~3,575)	ishrinks as screens extract

Component

Built

Tested

Cut over

Old code removed

Food screens (~1,340 lines) — BENCHMARK cluster

C2Uiiterating — built, now in a build ⇄ test loop — C2U transcribing food screens (TIMED benchmark); reads via parity map; writes dispatch-only

C2Uigated PASS; part of C2U solo UI sprint

Workout screens (~1,450)

C2Uiiterating — built, now in a build ⇄ test loop — C2U: workout screens PASSED (boundary-clean, dispatch-only)

C2Uiboundary guard green; gated

Voice/main + chat screens (~2,500)

C2Uiiterating — built, now in a build ⇄ test loop — C2U: main-voice-shell UI transcribed + gated (24/24 boundary, dispatch-only)

C2Uiboundary green

WorkLog screens (~1,050)

iDeferred per Howard 2026-07-03 — worklog UI transcription parked; revisit after the core modules

Notes / Todos / Lists / Settings screens (~1,320)

C2Uiiterating — built, now in a build ⇄ test loop — C2U: todos+lists+notes+settings screens all PASSED (16/16 boundary, dispatch-only)

C2Uiboundary guards green; gated

Food Lab + admin/debug panels (~3,060)

C2Uiiterating — built, now in a build ⇄ test loop — C2U: food-lab + admin panels transcribed + gated (part of 22/22 branch)

C2Uiboundary green

Conversation UI (~700)

C2Uiiterating — built, now in a build ⇄ test loop — C2U: conversation UI transcribed + gated (part of 22/22 branch)

C2Uiboundary green

Main render + styles (~3,575)

ishrinks as screens extract

New spine (the ~2K shell that replaces App.tsx)

Component	Built	Tested	Cut over	Old code removed
New shell — spec/design	ispec complete: FABLE_NEW_SHELL_SPEC_20260703.md (Claude-as-Fable) — transcription rule, manifest, phases, acceptance
New shell — build (App shell + root state, ~5,500 lines)	C1Siiterating — built, now in a build ⇄ test loop — P1 CLOSED 2026-07-03: F2 durable-write DB-verified (real todo_items row via new shell on sim). Full chain proven. P2 parity transcription underway	C1SiP1 exit met (boots+routes+navigates + REAL DB write, DB-verified); P2 parity next	iswap = differential-replay proof (old vs new shell)	iold App.tsx parked as reference at swap; never line-deleted

Component

Built

Tested

Cut over

Old code removed

New shell — spec/design

ispec complete: FABLE_NEW_SHELL_SPEC_20260703.md (Claude-as-Fable) — transcription rule, manifest, phases, acceptance

New shell — build (App shell + root state, ~5,500 lines)

C1Siiterating — built, now in a build ⇄ test loop — P1 CLOSED 2026-07-03: F2 durable-write DB-verified (real todo_items row via new shell on sim). Full chain proven. P2 parity transcription underway

C1SiP1 exit met (boots+routes+navigates + REAL DB write, DB-verified); P2 parity next

iswap = differential-replay proof (old vs new shell)

iold App.tsx parked as reference at swap; never line-deleted

Infrastructure & tooling (non-module workstreams)

Component	Built	Tested	Cut over
Gate-runner (runtime replay + DB-verify cert plane)	icertified overnight	icaught + confirmed a real bug	iBLOCKED on Howard: Mini needs the Developer-Tools TCC grant for Claude (simctl blocked). Build done+staged; runs the moment it clears.
slice-gate.mjs (automated per-slice code review)	iGrok-built, Claude-gated; already issued real verdicts		iin use
gate-dispatcher (watches branches, auto-invokes gates)	iC1O gate-dispatcher watcher PASSED (3763724d0, 4/4); loop-arm deferred to A1
Wake-bus (cross-machine poke/wake, Supabase Realtime)	C1OiStep 2 listener + tests + model probe + schema proposal + dry-run log BUILT (unarmed). Pending: A1 gate of schema/dry-run, then Howard authorizes (a) live wake_requests table apply (b) standing-arm.
Conversational corpus (180 rows, 4 types, core/stretch tiers)	iGrok-built, Claude-gated, landed	ivalidator clean; content-gated
Harness extensions for conversational types (2-turn seeding, ask/no-write scoring)	iR0 normalization layer gated PASS (typed_v2 CLARIFY/CA/DECLINE/OTHER -> expected fields + queue context threading)	C2Tirunning 180-row baseline scorecard on C2T sim
Full-corpus BASELINE bank (old-shell side of the swap diff)	A2iqueued after shakedown cert
Differential-replay tooling (old vs new shell diff)	A1iLane-H; after baseline banks
safe_food_fast_path_match widening (food coverage fix)	C2iiterating — built, now in a build ⇄ test loop — read-only proof staged (ba2d63868) — Claude gating	C2iv1 proven read-only: coverage +10 (14→24/66) AND false-admits improve (11→6, blocker-risk 0); v2 iterating before Howard decision	iships STAGED for Howard
Anti-drift fitness fns wired into spine CI	itools exist (ratchet/sole-writer/boundary-lint); CI wiring awaits Howard's call
Fleet reporter (C2R on Codex-2) — polls members q15m, drift-detection vs ledger	irelocated C1R→C2R per token-pool rule; C1R archived by Howard	ilive+registered, sole reporter	istanding

Component

Built

Tested

Cut over

Old code removed

Gate-runner (runtime replay + DB-verify cert plane)

icertified overnight

icaught + confirmed a real bug

iBLOCKED on Howard: Mini needs the Developer-Tools TCC grant for Claude (simctl blocked). Build done+staged; runs the moment it clears.

slice-gate.mjs (automated per-slice code review)

iGrok-built, Claude-gated; already issued real verdicts

iin use

gate-dispatcher (watches branches, auto-invokes gates)

iC1O gate-dispatcher watcher PASSED (3763724d0, 4/4); loop-arm deferred to A1

Wake-bus (cross-machine poke/wake, Supabase Realtime)

C1OiStep 2 listener + tests + model probe + schema proposal + dry-run log BUILT (unarmed). Pending: A1 gate of schema/dry-run, then Howard authorizes (a) live wake_requests table apply (b) standing-arm.

Conversational corpus (180 rows, 4 types, core/stretch tiers)

iGrok-built, Claude-gated, landed

ivalidator clean; content-gated

Harness extensions for conversational types (2-turn seeding, ask/no-write scoring)

iR0 normalization layer gated PASS (typed_v2 CLARIFY/CA/DECLINE/OTHER -> expected fields + queue context threading)

C2Tirunning 180-row baseline scorecard on C2T sim

Full-corpus BASELINE bank (old-shell side of the swap diff)

A2iqueued after shakedown cert

Differential-replay tooling (old vs new shell diff)

A1iLane-H; after baseline banks

safe_food_fast_path_match widening (food coverage fix)

C2iiterating — built, now in a build ⇄ test loop — read-only proof staged (ba2d63868) — Claude gating

C2iv1 proven read-only: coverage +10 (14→24/66) AND false-admits improve (11→6, blocker-risk 0); v2 iterating before Howard decision

iships STAGED for Howard

Anti-drift fitness fns wired into spine CI

itools exist (ratchet/sole-writer/boundary-lint); CI wiring awaits Howard's call

Fleet reporter (C2R on Codex-2) — polls members q15m, drift-detection vs ledger

irelocated C1R→C2R per token-pool rule; C1R archived by Howard

ilive+registered, sole reporter

istanding

Ask A1

VoiceCoach - Refactor + Fleet Dashboard

Fleet

Status Legend

Engine (utterance pipeline)

Modules (utterance domains)

UI screens — parallel transcription for the new shell (de-serialized; read-only App.tsx, new files only)

New spine (the ~2K shell that replaces App.tsx)

Infrastructure & tooling (non-module workstreams)