STATE.md: Truth-Everywhere Part 2 complete (all 8 items fingerprinted) + C4

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Kev
2026-07-18 01:56:45 -04:00
parent a8e383e7e7
commit b354d1d088
+14 -1
View File
@@ -1,5 +1,18 @@
# VYNDR — STATE OF THE WORLD
### As of `71e35e9` (main, deployed + fingerprint-verified live), 2026-07-17. This file opens every future session. Update it when a train ships.
### As of `a8e383e` (main, deployed + fingerprint-verified live), 2026-07-18. This file opens every future session. Update it when a train ships.
## TRUTH-EVERYWHERE PART 2 — COMPLETE (2026-07-18, on main, all 8 items fingerprinted)
Standing law: NO FABRICATED DATA ANYWHERE (UI, marketing, demos, examples). If it renders a number, it's from the DB or it doesn't render. Live data IS the better demo.
- **Item 1 — Verb law** (`66d52a9`): one verb, READ (never SCAN), swept every user-visible surface (FAB/nav/tier/ticker/upgrade). `tests/unit/verbLaw.test.js` enforces. Ticker "slate read" text flips at the next snapshot cron.
- **Item 2 — Founder counter** (`41fc2b9`): real active-paying-founder count from Stripe-synced user_profiles, cached; HIDES when unavailable. Live: **1/100** (was fabricated 47/100).
- **Item 3/4 — Reports**: full fabricated-data inventory + pricing claim audit (LIVE/PARTIAL). Alt-ladder/Kelly/correlation are LIVE in the engine; scratch-cascade + real-time-feed are PARTIAL.
- **Item 5 — Daily hero prop** (`9b9aab4`): live RULE = largest |modelconsensus| gap, A/B only. Shows book line vs VYNDR model + real timestamp. Empty→most-recent-real; nothing→hidden. NO static Jokic fallback. `heroPropService` + `/api/hero-prop`. Live: Erica Wheeler 9.5 vs 14.2.
- **Item 6 — Desk showcase** (`cb3237c`): real alt-line ladder from the snapshot (edge guarded on small-line artifact); φ = real same-team correlation for two real legs or hidden; Kelly tile REMOVED (no odds → can't size honestly). `deskShowcaseService` + `/api/desk-showcase`. Live: Vandersloot ladder.
- **Item 7 — Accuracy → TRUTH** (`89a2977`): public 30D accuracy (`/api/accuracy` + `/api/ledger/accuracy`) now reads the CLEAN Postgres ledger aggregate (getModelAggregate/getAccuracyView, model_value>0), not the Redis outcome log. Number moved 67%→61%. **BEAT CLOSE suppressed** at source (`clvCaptureReliable()`) — see C4.
- **Item 8 — Blog** (`3b7a1f5`,`a8e383e`): wired the 5 REAL articles (content lives in `web/content/articles` — the runtime content root, NOT repo-root content/) with honest dates + markdown render + OG. Backdated 2026-03-22 orphan deleted + 301. Live: 5 articles serve.
### 🔴 C4 — CLV CAPTURE BROKEN (open, backend)
`captureClosing` re-records the LOCKED line as "closing" (closing_line == locked_line on the whole sample) → CLV is flat → BEAT CLOSE = a measured-wrong 0%. Suppressed on all public surfaces until fixed. Full write-up + fix plan: `specs/audit-data/clv-capture-broken.md`. Flip `CLV_CAPTURE_RELIABLE=1` after C4 verifies real closes.
## PHONE AUDIT PUNCH LIST — COMPLETE (2026-07-17, on main, deployed, fingerprinted)
Kev's 35-screenshot real-390px audit (signed-out + Desk signed-in). Worked top-down, fingerprinted the batch (compare "on current form" literal live in served chunk off `71e35e9`). Test baseline: 268 suites / 3239.