60 · Audit · Fable plan

/60-v2 — Data Wiring Audit & Plan

Verdict: the standalone /60-v2 page is architecturally wired correctly. Every data surface calls the right tool through the browser-safe path (/api/realtime-tool → mcp_call_tool → toolResult, authed by the user's Supabase session). The reason it looks empty is environment, not wiring: /60-v2 is deployed on staging, whose Supabase has £0 pipeline and QA-synthetic test meetings for this user. Production holds the real data (404 open deals, £1.06M, real analysed calls) — verified: every tool returns full data on prod.

So the #1 action is: get /60-v2 onto production, where the real CRM lives. Then close the handful of genuine wiring gaps below.

1. Surface-by-surface audit

Surface	Page calls	Path	Wired?	Prod data	Staging data
Chat answer	`t2.ask` / `askViaStream` → `/api/chat`	Session JWT	✅	✅ real	✅ (sparse)
Morning brief	`loadBrief` → search_meetings + pipeline_summary + list_tasks	realtime-tool	✅	✅	empty (£0)
Judgement deck	`loadBrief` deck (tasks + pipeline-risk + commitments)	realtime-tool	✅	✅ (from 404 deals)	empty
Meeting canvas	`loadMeeting` → search_meetings + intelligence + transcript	realtime-tool	✅	✅ real transcript	QA-synthetic
Ops table	`loadTable` → pipeline_summary / search_records / list_tasks	realtime-tool	✅	✅ (record-level via #91)	sparse
Approvals	`sixty_run_write_workflow`	realtime-tool	✅ built, ⚠️ unverified on prod	needs a real write test	—
Pipeline card	pipeline_summary	realtime-tool	✅	✅	£0

Genuine gaps (not data-env):

Voice — mic shows a "voice is in the /60 app" toast; the real /api/realtime-session WebRTC isn't wired into the standalone page.
Status pill — static 60 · V3 instead of a live gateway /health check.
History / credits — route-to-app / static ring instead of the prototype's in-page popovers.
Approval round-trip — built (mint → replay) but not yet exercised against a real prod write (safest: a task-create).

2. The plan

Phase 1 — Get `/60-v2` to production (unblocks ALL real data)

The page + route + scoped styles are additive and legacy /60 is untouched, so this is low-risk for the existing app. Two options:

A. Promote staging → main (PR #2402 lineage): carries every stream + /60-v2 commit. Broadest, but also rides the accumulated staging backlog (other teams' #2427/#2428, migrations, edge fns) — needs the migration/CI checks green first (the cron-category + dedupe fixes already landed; re-verify).
B. Cherry-pick the /60-v2 + stream commits onto a clean feat/60-v2-prod branch off main → PR → merge. Surgical, ships only our work to prod, avoids the backlog. Recommended for speed + safety. The prod gateway already has all fixes (brain 98/96, record tools, voice-context, toolResult passthrough) — verified deployed. So once the app route is on prod, app.use60.com/60-v2 renders real data immediately.

Phase 2 — Close the wiring gaps

Voice — wire useVoiceCallStarter/voiceCallStore + the realtime driver into Sixty60V2Page's voice bar (reuse the plumbing from #2422; the standalone page needs its own hook-up). Speak-the-headline is server-side (gateway spokenSummary) — already improved by voice-context (#90).
Status pill — poll gateway /health, show live tool count / connected state.
History + credits — port the prototype's in-page popovers (history from the conversation store, credits from the billing/credits endpoint).

Phase 3 — Verify every archetype on prod with real data

Drive each surface on app.use60.com/60-v2 (or a prod-session headless run) and confirm real rendering: brief + deck populated from 404 deals, meeting canvas with a real transcript, ops table with real deal rows, a real approval round-trip (task-create — never an external send), analytics/chart. Fix any mapper drift found (e.g. the JSON-leak class of bug, already fixed for meetings — audit the rest).

Phase 4 — Rollout

/60-v2 URL for the team (prod) → gather feedback.
Wire the org feature flag so V3 can become the default /60 per-org, then globally — retiring the separate URL.
Nightly eval SLO already guards the brain; add a light /60-v2 smoke to the canary.

3. Execution order

Phase 1B — cherry-pick /60-v2 to prod → real data live (fastest win).
Phase 3 — verify archetypes on prod, fix any mapper drift.
Phase 2 — voice, status pill, history/credits (parallelisable).
Phase 4 — rollout + make-default.

North star: app.use60.com/60-v2 shows Andrew's real pipeline, real meetings, real deck, working voice — pixel-matched to the prototype — then it becomes the default /60.

Prepared 7 Jul 2026 · Fable · audit verified live against prod + staging gateways.