vyndr/specs/HANDOFF-2026-08-08.md

VYNDR build handoff — 2026-08-08

Written from the repo, not from summary. Every claim below was grep- or run-verified at 71d3b7b. Where something could not be verified from the repo it says UNKNOWN.

Repo state

HEAD        71d3b7b78692c31ba0596ec88874986ff1b58116
gitea main  71d3b7b78692c31ba0596ec88874986ff1b58116   (in sync)
tree        clean — 0 modified/untracked
tests       4,539 passing / 4 skipped
web build   exit 0

Push remote is gitea, not origin. origin is GitHub and has no credential on this machine; every push must be git push gitea main. This cost twenty commits of false "no credentials" diagnosis — see docs/GIT-PUSH-DIAGNOSIS.md.

---

BUILD STATE per surface

DONE this arc — verified present

surface	file	lines
E1 movement strip	web/src/components/vyndr/MovementStrip.tsx	113
F9–F11 offseason hub shell	web/src/app/offseason/page.tsx + components/vyndr/OffseasonHub.tsx	21 + hub
E10 Report issue template	src/services/report/reportTemplate.js	169
E12 /report archive	web/src/app/report/page.tsx + components/vyndr/ReportArchive.tsx + src/routes/report.js	17 + comp + route
Content engine	src/services/content/contentEngine.js + 3 templates + cardRenderer.js	172 +
Content studio API + preview	src/routes/contentStudio.js + web/src/app/studio/page.tsx	175 + 131
Wave-D1 motion primitives	web/src/lib/motion.js (E17/E18/E27/E28)	100
Honest served grade	src/services/model/servedGrade.js	142
Archetype doctrine	specs/ARCHETYPE-TAXONOMY-DOCTRINE.md	53

Also DONE and verified: D1 glyph library (39 of 83 wired — every glyph that maps to a real archetype; colours match the registry exactly), A1 card token (--bg-1, 32 consumers), B1 boundary channel (--priced-out, 4 consumers), book comparison (mounted via GradeResultCard → scan/page), THE WIRE (vyndr/Ticker).

GATED — verified absent, with the gate named

surface	gate	verified
F5 article media	card-system reconciliation	ArticleHero/StatCallout/PullQuote = 0 files
E16/F8 share cards + crops	card-system reconciliation and the resolution tail has no share-card generation step	ShareCard.tsx has 0 importers
In-season hub IA	the content formula (social chat)	no spec exists anywhere
E9 calibration curve	model accrual — 0 eligible dates	CalibrationCurve = 0 files
E15 Price Triplet MODEL leg	model — EV layer	renders honest NO_MODEL
E2 BookChip tiles / E6 push-to-book	licensing / affiliate approval	every book enabled:false
E3 best-number crown	measurement — measured flat	CrownBadge = 0 files
E13 Scanner S6	another build order	—

There are no ungated Wave-2 targets left. The next move is a decision, not a build.

---

THE TWO OPEN DECISIONS

1. Card-system reconciliation — unblocks F5 and E16/F8 together

The content engine emits 1080×1350 SVG cards (src/services/content/cardRenderer.js). The design package defines five master sizes with layouts and rasterised PNGs in specs/design-reference/exports/: settle 1080×1350, story 1080×1920, square 1080, X 1200×675, record 1080×1350, article OG 1200×630.

I built the renderer without checking whether a designed card system existed. It did. They are a parallel invention — not wrong, but they must become one visual language before F5 (whose hero graphics are generated data visuals) or E16 can proceed.

The decision: does the content engine conform to the E16 masters, or do the masters absorb the engine's generator? Cheapest unblock on the board — it frees two gated items at once.

2. In-season hub information architecture

Vyndr Offseason.dc.html specifies an offseason hub, and its shell is built. Nothing specifies what that surface is in-season, or how content, articles, wire and the live slate share year-round navigation. Every component exists; the composition does not.

The spec is explicit that sport state lives in the sport tab (NFL · CAMP −5D) and never in a separate offseason tab — so this is not a new route, it is a mode. Gated on the social chat's content formula, which determines the recurring slots.

---

ACCRUAL CLOCK

champion marker: engine1@2026-08-07-fullwindow

  calibration_refit        0/10  WAITING     (~2 weeks)
  hits_factor_lift         0/10  WAITING     (~2 weeks)
  prior_verdict_reaudit    0/14  WAITING     (3+ weeks)
  rbi_lineup_slot_gate     0/14  WAITING     (3+ weeks)

blocked: no settled rows yet carry the repaired champion marker

FIRST TRIGGER: 10 eligible calibration dates → calibration re-fit. Then, in order: hits factor lift → prior verdict re-audit → rbi lineup-slot gate.

Thresholds are ATTEMPT floors, not TRUST floors. Reaching 10 dates means the re-fit can be measured, not that it is trustworthy. A 10-date map is thin, deploys PROVISIONAL with auto-demotion, and its interval will be wide.

No measurement on reconstructions of the retired forecast. Anything requiring settled data waits. src/services/model/reAuditEligibility.js enforces it — eligibility is a version marker, counted in DATES not rows, and a mixed table counts only the repaired rows.

---

STANDING DOCTRINE

Truth Law. No fabricated data anywhere — including designer sample data. The design files' numbers (Nabers 1,120.5, Wembanyama +420→+330, Nº 128, DAY RECORD 9–4) are a spec for what a live feed renders, never content to paste. Tests assert none of it ships. Number(null) === 0 is the classic breach; note its inverse also bites — zero is a real fact ("0 cleared B+" is our most honest post).

83-glyph four-sport taxonomy. A glyph renders ONLY where its archetype is modeled and proven. 39 wired; 44 dormant slots for WNBA/NBA/Soccer. Wiring them would mean inventing 44 archetypes to consume artwork — decoration presented as classification, forbidden. DUAL THREAT and PAINT BOSS are modeled archetypes with no mark: the mirror gap, flagged to design.

Repaired champion. The forecaster read res.last10 — ten games — as its season rate. Fixed to the full season log with recency weight 0.40 → 0.20. Resolution: hits 0.00251 → 0.00817. Every factor verdict in the programme was measured against the broken baseline and may deserve re-audit — direction unknown, not pre-priced.

Calibration is WITHDRAWN. CALIBRATION_DEPLOYED = []. The maps were fitted on the retired forecast; refitting today would refit it again. Served p_win is repaired-champion raw.

Grade doctrine. A+, A, A- are unissuable — structurally, not rarely. Ceiling is B+ at 0.663 realized against a 0.6005 baseline. Bands that cannot be separated from the baseline SAY so. The served letter derives from p_win; engine1.grade is preserved as engine_grade and read by no serving code.

Two guards, both in CI:

• renderReachability.test.js — a promised field must trace payload → adapter → component → mounted. Built-but-unread killed three surfaces before this.
• baseRateWindow.test.js — no fixed N may stand in for a season. Six paths across two sports fell to that class.

Compose, don't fork. The recurring failure this arc: building beside an existing thing instead of on it (cards vs E16, MovementStrip nearly vs gradeShift, reportTemplate nearly vs newsletterService). Check for the existing implementation first — twice it was already there.

---

PARALLEL WORKSTREAM SEEDS

A. Social strategy chat

VYNDR's content engine ships posts from real data (src/services/content/, three templates, Truth-Law enforced — an unbacked token refuses to render). Review /studio output and specs/CONTENT-ENGINE.md. Produce: the content formula/calendar (what posts recur, on what cadence), and the card-format decision — the engine's 1080×1350 renderer vs the five designed masters in specs/design-reference/exports/. That decision unblocks F5 article media and E16/F8 on the build side. Voice: specs/VOICE.md governs; the engine templates encode a sharper register — decide whether that becomes the house voice.

B. Multi-sport archetype build

VYNDR has archetype registries for MLB (15), NBA (15), Soccer (6), WNBA (5) — but only MLB batters is MODELED; the rest have no proven factor model, so their glyphs stay dormant per specs/ARCHETYPE-TAXONOMY-DOCTRINE.md. 44 of 83 designed glyphs await their sport. Pick a sport, build its archetype classifier from real features, and take it through the two-part gate (src/services/model/factorGate.js — must MOVE the prediction and improve out-of-sample Brier, cumulative-Bonferroni corrected). Prerequisite: that sport needs settled outcomes. WNBA settles via ESPN box scores; NBA and Soccer do not settle at all yet, which is the real first task there.

---

Where to start in the next chat

1. Take one of the two open decisions (card system is cheapest — unblocks two).
2. Or run a parallel workstream seed above.
3. Do not start model work — it is date-blocked and will stay so for ~2 weeks.
4. Verify before building: this arc corrected its own board four times. git grep beats any summary, including this one.