Files

T

builtbykev e70c02aca0 report: incomplete-surface build triage; boundary tag NOT applied (no promotion exists)

Report-only. Nothing built, wired, tagged, or removed.

PART 1 — the recalibration boundary was NOT written. The build order asks to tag
grades pre/post edge-shading at "the true promotion timestamp"; there is no such
timestamp, and writing the marker would insert a fabricated model transition
into an append-only public record — the exact corruption the order exists to
prevent. Three independent production proofs:

  A. model_snapshots.code_sha — every sha that ran the pipeline in the last 5
     days is a documented commit from this session (f3bf300 currently live,
     then f310608, 9b5235c, b8ee216, afb56b1, 3592aba, 914a057), all stamped
     model_version engine1@2026-07-20. No promotion commit exists.
  B. Daily A-family share 07-24..07-30: 0.0, 0.0, 0.0, 0.0, 1.0, 1.6, 0.0 —
     flat at zero, no step change on any date. A 92.9% re-letter under shading
     would have driven the board to ~79-93% A overnight.
  C. efficiencyShading has zero production importers; HEAD d54eca0, clean tree.

Phase 2 delta: neither 92.9% nor 43.6% is attested in any measurement here, and
no re-lettering occurred at any scale, so partial-slate-vs-full-board cannot
explain a gap that does not exist. The only measured numbers, on all 1250 rows
(the full board): 97.4% would change, 79.8% up, 79.0% A-family, and 0 of 1250
rows actually shaded — the hypothetical re-letter would have come entirely from
an unapproved grading-basis switch, which is why the flip was refused.

Measurement integrity needs nothing new right now: model_version already
separates the S64 eras and takeable tags are complete (1245/1250). The boundary
becomes a hard prerequisite the day a promotion actually ships.

PART 2 — build triage for 14 incomplete surfaces, classified from code with each
one's real dependency and honest size, sequenced into five waves: pure wiring
(/intelligence, /slip, /parlay, /marketplace after a copy honesty pass);
design-only gaps (Live tracking, Slip reader, Newsletter artboards);
self-contained builds (/compare, ShareCard host, /notifications); model-gated
(price triplet MODEL leg and the calibration board both need p_win vs
fair_prob — building either on edge_pct would re-ship the retired 620% lie);
and sport-boundary/quota-gated (/soccer blocked on odds-api 0/500, /system,
Offseason).

Matrix corrections: Live tracking, Newsletter and Slip reader are built/live/
honest and marked incomplete only on the DESIGN column; /parlay is
drawer-reachable, not unreachable; /compare is already honest. No live surface
is showing fabricated data today.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QJs13VsyiSKYQP6rj3NNmc

2026-07-31 01:49:55 -04:00

8.5 KiB

Raw Blame History

VYNDR — INCOMPLETE-SURFACE BUILD TRIAGE + RECALIBRATION-BOUNDARY FINDING

Report-only, 2026-07-31. Nothing built, wired, tagged, or removed.

PART 1 — THE BOUNDARY TAG WAS NOT APPLIED. THE PROMOTION NEVER HAPPENED.

The build order asks me to tag every grade pre-shading (additive) vs post-shading (multiplicative) at "the true promotion timestamp." There is no such timestamp. Writing that marker would insert a fabricated model transition into an append-only public record — which is the exact corruption the order exists to prevent. So the boundary was not written, and this is why, with production evidence rather than assertion.

Evidence A — the deployed code, from `model_snapshots.code_sha` (the pipeline stamps its own sha)

Every sha that ran the pipeline in the last 5 days:

code_sha	model_version	window
`f3bf300` (C1 takeable-floor report)	engine1@2026-07-20	07-30 22:00 → 07-31 03:02 (current)
`f310608` (top-graded fingerprint docs)	engine1@2026-07-20	07-30 03:00 → 19:02
`9b5235c`, `b8ee216`, `afb56b1`, `3592aba`, `914a057`	engine1@2026-07-20	07-26 → 07-30

Every one is a documented commit from this session's history. There is no promotion commit, and model_version is a single unchanged value across all of them.

Evidence B — the grade distribution shows no step change, ever

A flip re-lettering 92.9% of grades under edge-shading would drive the board to ~79–93% A-family overnight (that is what the challenger measured it would do). Actual daily A-family share:

date	07-24	07-25	07-26	07-27	07-28	07-29	07-30
n	20	25	44	26	98	62	91
% A-family	0.0	0.0	0.0	0.0	1.0	1.6	0.0

Flat at zero throughout. No discontinuity on any date.

Evidence C — the challenger is imported by nothing

efficiencyShading has zero production importers; it remains measured-never-served. HEAD is d54eca0, clean tree.

Phase 2 — the 92.9% vs 43.6% delta, explained with row counts

Neither figure is attested anywhere in this project's measurements, and no re-lettering occurred at any scale. The only measured numbers, on all 1,250 public ledger rows (full board, not a partial slate — so "partial-slate vs full-board" cannot explain a gap that does not exist):

grades that would change if promoted: 1,217 / 1,250 = 97.4%
direction: 79.8% up, 17.6% down → 79.0% A-family vs the champion's 0.2%
rows actually shaded by efficiency: 0 / 1,250 (96.5% of markets unscored → f=1; the one scored market present is the f=1 anchor)

So the hypothetical re-letter was never 43.6% or 92.9%, and it would have been driven entirely by switching the grading basis to edge-vs-fixed-bar — not by efficiency shading, which is inert on this board. That unapproved basis change is precisely why the flip was refused.

What measurement integrity actually needs right now: nothing new.

model_version already separates the S64 eras (engine1@2026-07-20 / pre-retention-unknown), and takeable tags are complete (1,245/1,250, takeable_floor on all, one floor −160; the 5 untagged have no locked price = honest NULL). The boundary becomes a hard prerequisite on the day a promotion actually ships — and config/modelEras.js is its home.

PART 2 — INCOMPLETE-SURFACE BUILD TRIAGE (all get built; nothing cut)

Classified from the code, not the matrix.

#	Surface	Real state (cited)	Real dependency	Size
1	`/intelligence`	BUILT-NOT-WIRED. `app/intelligence/page.tsx`, 160 lines, real data, 0 placeholders, 0 nav links	none — it works	small wire (1 nav entry)
2	`/slip`	BUILT-NOT-WIRED. `app/slip/page.tsx`, 526 lines, live, 0 nav links	none — OCR + `/api/slips/parse` live	small wire
3	`/parlay`	BUILT, drawer-only. 340 lines; reachable via `#parlay`, 0 direct nav links	none	small wire (+ a design artboard)
4	`/marketplace`	BUILT-NOT-WIRED. 121 lines, waitlist/interest capture (API access etc.), 0 nav links	⚠️ honesty check: it advertises unbuilt capabilities — must read as interest capture, never as available	small wire after copy check
5	`/soccer`	BUILT-NOT-WIRED. 414 lines, real fetch → `/api/odds/soccer/:league`	🔴 odds-api quota (0/500) — soccer props blocked on QUOTA, not code. Also needs a soccer settled-result feed or it grades and never settles	medium (blocked upstream)
6	`/notifications`	DESIGNED-NOT-BUILT. 14 lines, `RouteStub` ×2	a notification store + delivery (email/push) + user prefs	real build
7	`/compare` (H2H)	BUILT-PLACEHOLDER→honest. 27 lines, in-dev state (fabrication already removed)	two-player fetch over `/api/stats/player` (exists)	medium
8	`/system`	DESIGNED-NOT-BUILT. No `app/system/` dir; prod 404. `System.dc.html` exists	decide vs `/intelligence` overlap — may be the same surface designed twice	real build
9	Offseason hub	DESIGNED-NOT-BUILT. No dir; prod 404. Full `Offseason.dc.html`	futures/news feeds (partly live in `FuturesBoard`/`NewsWire`)	real build
10	Article media / `ShareCard`	BUILT-NOT-WIRED (dead). 0 real importers	a share/OG surface to host it; `opengraph-image.tsx` already live	medium
11	Calibration / edge board	BUILT-NOT-WIRED (correctly hidden). `MobileEdgeBoard`, 0 importers	🔴 a REAL edge feed — i.e. `p_win` vs `fair_prob`. Cannot ship on `edge_pct` (price-free, line-scaled)	medium, gated on the model
12	Price triplet MODEL leg	BUILT, not PRODUCING. `NO_MODEL` honest-absent renders correctly	🔴 the EV layer must emit `p_win`/`ev_pct`/`model_odds` on served grades	medium, model work
13	Live tracking	BUILT + LIVE + HONEST. Design is the only gap	a design artboard	design task
14	Newsletter	BUILT. Send is internal-only	Listmonk env (`CANNOT DETERMINE` externally) + a design artboard	small (config + design)

2.6 — matrix corrections (accuracy, not cutting)

Live tracking (#15) is marked ❌ only because DESIGNED=NO. It is built, wired, live and honest — the gap is an artboard, not code. Same for Newsletter (#21) and Slip reader (#22) on the design column.
/parlay is not "unreachable" — it is drawer-reachable; the gap is a direct link.
/compare is already honest (the Jokić/Wembanyama fabrication is gone); only the real build remains.

PHASE 2 — ORDERED BUILD PLAN (dependency-sequenced)

Wave 1 — zero dependencies, pure wiring (hours):

/intelligence nav link · 2. /slip nav link · 3. /parlay direct link ·
/marketplace link after an honesty pass on its copy.

Wave 2 — design-only gaps (no code): 5. Artboards for Live tracking, Slip reader, Newsletter; promote the System bundle's "PARLAY BUILDER" to a named Parlay-lab artboard. 6. Confirm Listmonk env → Newsletter LIVE.

Wave 3 — self-contained builds: 7. /compare real two-player fetch (deps already live) · 8. ShareCard host surface · 9. /notifications (needs a store + delivery — the largest of the three).

Wave 4 — gated on the MODEL, do not start before it: 10. Price triplet MODEL leg and 11. Calibration/edge board both need a real price-aware edge (p_win vs fair_prob). Building either on edge_pct would re-ship the retired 620% lie.

Wave 5 — gated on the SPORT BOUNDARY + upstream quota: 12. /soccer — blocked on odds-api quota (0/500), and needs a settled feed. 13. /system — first decide whether it duplicates /intelligence. 14. Offseason hub.

Placeholder honesty during the build (Truth Law)

Only two live surfaces carry anything placeholder-ish, and neither currently displays fabricated data: /compare (honest in-dev copy) and /marketplace (advertises unbuilt capability — flagged for a copy pass so it reads as interest capture). MobileEdgeBoard's placeholder edge feed is already unrendered. No live surface is showing fabricated data today.

8.5 KiB Raw Blame History Unescape Escape