builtbykev
c2c7abbb65
Challenger only. Champion grade byte-identical (verified by diff). Nothing promoted, no live grade re-lettered, no ledger row deleted or re-settled. BUILT src/services/challengers/efficiencyShading.js (measured-never-served): adjusted_edge = raw_edge * f(efficiency); grade = band(adjusted_edge) against ONE fixed bar (A+>=10, A>=5, B>=3, C>=1, D>=0, F<0) that never moves. f(e) = E_SOFTEST/e bounded to (0,1] — soft markets intact (never amplified), sharp shaded toward but not past zero, unscored -> f=1 and FLAGGED. A fence test asserts no production grade path imports it. Cross-market behaviour is unit-proven: the same raw 6% edge grades A in soft mlb:total_bases and B in sharp nba:points. MEASURED on 1250 live ledger rows — Phase 2.5's answer is NO, the flooding is not gone: challenger 79.0% A and 80.9% A/B (MLB 93.4% A) vs champion 0.2% A. TWO findings explain why, and they are the point of the order: 1. The shading is a NO-OP on the live board: rows_actually_shaded = 0 of 1250. 96.5% of rows are UNSCORED (f=1), and the one scored market present (mlb:total_bases) is the anchor so its f is 1.0 by construction. mlb:strikeouts and nba:points do not appear in the ledger at all (our basketball is wnba, not nba). Challenger vs baseline: 0 rows changed. 2. Placement was never the bug — the INPUT SCALE is. Against a fixed 5% bar the RAW edge already clears A on 100% of MLB doubles, 89.6% of hits, before any shading. MLB median raw edge is 60%, twelve times the bar. Decisive test: apply the sharpest score in the spec (f=0.647) to EVERY row — the maximum the design permits — and 75.8% still clear A (MLB 91.7%). Since f is bounded <= 1, no achievable shading can close a 12x overshoot. Moving the multiply from the threshold to the edge does not change the outcome. This is edge_pct behaving as the 2026-07-29 diagnosis described: a price-free (proj-line)/line gap whose scale is a function of line size. It is not a betting edge, so no fixed betting-edge bar is meaningful against it. 2.6 efficient-market over-suppression: CANNOT DETERMINE — zero live rows are shaded, so there is no efficient market in the data to over-suppress. Phase 3: takeable tagging was completed in the previous order (migration 034, 1246/1254 rows) and is not repeated. The model-version boundary is again NOT applied: nothing promoted, so no boundary exists. Unblocking needs the input replaced, not the multiply moved: p_win vs fair_prob (both already computed) instead of edge_pct, plus scores FIT from our own record for the markets we actually grade. Floor: 313 suites / 3899 tests green (9 new), web build exit 0. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QJs13VsyiSKYQP6rj3NNmc
115 lines
6.4 KiB
Markdown
115 lines
6.4 KiB
Markdown
# SPEC — EDGE-SIDE EFFICIENCY SHADING CHALLENGER (built + measured; NOT promoted)
|
||
2026-07-31. Champion grade UNCHANGED (byte-identical, verified by diff). Push scoring untouched.
|
||
|
||
---
|
||
|
||
## REVIEW ZERO
|
||
|
||
- **0.1 raw edge — EXISTS.** `analyzeViaEngine1.edgePctFor` = `((proj − line) / line) × 100`,
|
||
signed by direction; persisted as `ledger_entries.edge`.
|
||
- **0.1 threshold application — DOES NOT EXIST.** The champion grade is `engine1`'s additive
|
||
factor index; there is no edge-vs-threshold step on the live path.
|
||
- **0.1 the flat additive ±0.05 efficiency nudge — DOES NOT EXIST** (third re-check). No
|
||
`efficiency` reference anywhere in `src/services/intelligence/` or `gradeAdapter`. **There is
|
||
nothing to remove from the challenger path.**
|
||
- **0.2 fixed base thresholds — DO NOT EXIST live.** So the challenger DEFINES one and it never
|
||
moves: **A+ ≥ 10, A ≥ 5, B ≥ 3, C ≥ 1, D ≥ 0, F < 0**, identical for every market. (The spec's
|
||
per-sport MLB-5%/NBA-7% variant is deliberately NOT used — a bar that differs by sport is a
|
||
moving bar, which this design forbids.)
|
||
- **0.3 efficiency scores — 5 exist in spec, and they cover almost nothing we grade.**
|
||
`nfl:passing_yards .85 · nba:points .80 · mlb:strikeouts .60 · mlb:total_bases .55 ·
|
||
nba:role_player .55`. Of the **11 live ledger markets**, only `mlb:total_bases` matches.
|
||
Unscored markets default to **f = 1** and are FLAGGED.
|
||
|
||
## THE BUILD (`src/services/challengers/efficiencyShading.js`, measured-never-served)
|
||
|
||
adjusted_edge = raw_edge × f(efficiency)
|
||
grade = band(adjusted_edge) ← ONE fixed bar, never moved
|
||
|
||
`f(e) = E_SOFTEST / e`, bounded to **(0, 1]**. At the softest scored market (0.55) `f = 1.0` —
|
||
the edge is left **intact, never amplified** (the guardrail). Sharper markets shade down
|
||
monotonically: `f(0.60)=0.917`, `f(0.80)=0.688`, `f(0.85)=0.647`. Unscored → `f = 1`, flagged.
|
||
Absent edge → `null` (absent, never graded F). A fence test asserts **no production grade path
|
||
imports this module**.
|
||
|
||
**Cross-market behaviour works as designed** (unit-proven): the same raw **6%** edge grades
|
||
**A** in soft `mlb:total_bases` and **B** in sharp `nba:points`.
|
||
|
||
---
|
||
|
||
## PHASE 2 — THE MEASUREMENT, AND IT IS A **NO**
|
||
|
||
### 2.5 "Confirm the naive-formula pathology is GONE" → **IT IS NOT GONE.**
|
||
|
||
| population | n | challenger %A | challenger %A/B | champion %A | champion %A/B |
|
||
|---|---|---|---|---|---|
|
||
| MLB | 696 | **93.4** | **93.7** | 0.0 | 57.9 |
|
||
| WNBA | 554 | 61.0 | 64.8 | 0.5 | 58.3 |
|
||
| **TOTAL** | **1,250** | **79.0** | **80.9** | 0.2 | 58.1 |
|
||
|
||
**80.9% A/B — the flooding the fix was meant to remove is still there** (the naive formula's
|
||
reported 86% A/B is not meaningfully improved on).
|
||
|
||
### TWO FINDINGS THAT EXPLAIN WHY, AND THEY ARE THE POINT OF THIS ORDER
|
||
|
||
**(1) The shading is a NO-OP on the live board. `rows_actually_shaded = 0` of 1,250.**
|
||
**96.5% of rows (1,206) are UNSCORED** → `f = 1`. And the one scored market present
|
||
(`mlb:total_bases`) is the *anchor*, so its `f` is 1.0 by construction. `mlb:strikeouts` and
|
||
`nba:points` do not appear in the ledger at all (our basketball is **wnba**, not `nba`).
|
||
**Challenger vs baseline: 0 rows changed.** The design cannot act on a board it has no scores for.
|
||
|
||
**(2) Placement was never the bug — the INPUT SCALE is.** Against a fixed 5% bar, the **raw**
|
||
edge already clears A on 100% of MLB doubles, 100% stolen_bases, 89.6% hits, 88.6% total_bases —
|
||
**before any shading at all**. MLB's median raw edge is **60%**, twelve times the bar.
|
||
|
||
> **The decisive test: apply the SHARPEST score in the entire spec (f = 0.647) to EVERY row —
|
||
> the maximum shading the design permits — and 75.8% still clear the A bar** (MLB 91.7%).
|
||
> Since `f` is bounded ≤ 1, **no achievable shading can fix this.** Moving the multiply from the
|
||
> threshold to the edge does not change the outcome, because a bounded multiplier cannot close a
|
||
> 12× overshoot.
|
||
|
||
This is `edge_pct` doing exactly what the 2026-07-29 diagnosis said it does: it is a price-free
|
||
`(proj − line)/line` gap whose scale is a **function of line size**, so on 0.5-lines it explodes
|
||
(median 60, max 620). **It is not a betting edge, so no fixed betting-edge bar is meaningful
|
||
against it.**
|
||
|
||
### 2.6 efficient-market over-suppression → **CANNOT DETERMINE.** Zero live rows are shaded, so
|
||
there is no efficient market in the data to over-suppress. The failure mode is untestable here.
|
||
|
||
### 2.7 cross-market behaviour → **works in unit test, ZERO live effect** (see finding 1).
|
||
|
||
---
|
||
|
||
## PHASE 3 — TAGGING
|
||
|
||
- **Takeable tagging: ALREADY DONE** in the previous order (migration 034; 1,246 of 1,254 rows
|
||
tagged, 781 takeable / 465 below floor, `null_despite_price = 0`, floor −160 recorded per row
|
||
and labelled policy-not-derived). **Not repeated** — re-running would be a no-op.
|
||
- **Model-version boundary: DELIBERATELY NOT APPLIED, again.** It marks pre/post a scaling
|
||
change. **Nothing promoted, so no boundary exists** — stamping one would record a model
|
||
transition that never happened. `config/modelEras.js` is its home the day one promotes.
|
||
|
||
---
|
||
|
||
## WHAT WOULD ACTUALLY UNBLOCK THIS
|
||
|
||
1. **Replace the input, not the multiply.** A bar-based grade needs a quantity that is a real
|
||
edge — model probability vs the de-vigged market price. That is `p_win` vs `fair_prob`, both
|
||
already computed and stored. `edge_pct` cannot carry a fixed bar at any shading.
|
||
2. **Fit the scores from our own record**, per market we actually grade — 9 of 11 live markets
|
||
have no score, and inventing 9 numbers would be fabricated calibration.
|
||
3. **Then** re-run this challenger; the mechanic itself (bounded, edge-side, fixed bar) is sound
|
||
and already built.
|
||
|
||
## VERIFY
|
||
Champion byte-identical (diff empty across `intelligence/`, `gradeSlateService`, `gradeAdapter`,
|
||
`snapshotService`, `grade_thresholds.json`). Fixed bar never moved — efficiency multiplies the
|
||
edge only, and a test asserts shading can only lower a grade. **313 suites / 3,899 tests green**
|
||
(9 new), **web build exit 0**. Nothing promoted, no live grade re-lettered, no ledger row
|
||
deleted or re-settled. Auth-gated visuals → Chrome audit; nothing visual changed.
|
||
|
||
## TAGS
|
||
VERIFIED: the no-op (0 rows shaded, 96.5% unscored); the flooding (80.9% A/B) and that max
|
||
shading still leaves 75.8% clearing A; champion untouched. **CANNOT DETERMINE: efficient-market
|
||
over-suppression (no shaded rows exist).** BLOCKED: promotion — the input scale must change first.
|