Player projections the model has generated and graded against real match outcomes, by date range — with how often each stat actually occurred (graded events). See the per-game projections vs actuals.
How the model's edge has performed
as of 2026-06-29Higher edge → higher tier. Edge is the gap between our model's probability and the market price. We rank value from the data and surface the top tiers — data, not tips.
Two independent evidence points, not a like-for-like A/B — different competitions and different prices. OVERS only · markets shots_opta/SOT/assists/fouls · flat 1u.
Leagues
PROJECTEDprojected · +15% assumed softness
n 37,545 settled overs · 2025-08-15 → 2026-05-24 · flat 1u
By tier — all markets
| Cut | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| All overs | 37,545 | 21.7% | 6.93 | -20.5% |
| ● Value (edge >= 12.5%) | 1,320 | 25.2% | 5.69 | +6.5% |
| ● Best (edge >= 25%) | 177 | 20.9% | 6.95 | +19.8% |
By tier — shots only
| Cut | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| All shots overs | 8,504 | 41.7% | 2.32 | -8.4% |
| ● Value (edge >= 12.5%) | 353 | 35.7% | 3.18 | +0.8% |
| ● Best (edge >= 25%) | 27 | 29.6% | 6.33 | +70.4% |
(Leagues profit concentrates in the top ~3.5% — the Best/Value tiers above. The broader edge deciles are all negative on these close prices, so the decile breakdown is omitted; the tier rows carry the story.)
Top-decile by market (positive only)
World Cup — group stage
ACTUALactual live run-up prices
n 3,639 settled overs · 2026-06-15 → 2026-06-26 · flat 1u
By tier — all markets
| Cut | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| All overs | 3,639 | 27.6% | 4.80 | -27.0% |
| ● Value (edge >= 12.5%) | 403 | 18.4% | 8.67 | +7.8% |
| ● Best (edge >= 25%) | 145 | 14.5% | 12.44 | +17.9% |
By tier — shots only
| Cut | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| All shots overs | 997 | 34.1% | 3.08 | -20.4% |
| ● Value (edge >= 12.5%) | 165 | 21.8% | 7.78 | +33.3% |
| ● Best (edge >= 25%) | 61 | 23.0% | 11.68 | +88.8% |
Edge deciles — ROI climbs with edge
Top-decile by market (positive only)
Leagues = projected at a conservative +15% softness (a forecast, not a settled record). World Cup = actual live run-up prices. Flat 1u stakes.
Game stats
Every game market with a model, a captured price and a settled actual, run through the same edge cut — leak-clean, graded at the real best-soft price, with fixture-clustered 95% CIs. Read the CI, not the point ROI.
Corners
ACTUALWC · leak-aware · no-arb two-way guard · fixture-clustered CI · n 365
| Cut | n | Win% | ROI | 95% CI |
|---|---|---|---|---|
| All (no-arb guarded) — canonical | 365 | 50.7% | +13.3% | [-4, +33] |
| ● Value (edge ≥ 12.5%) | 80 | 57.5% | +54.1% | [-12, +121] |
| ● Best (edge ≥ 25%) | 17 | 47.1% | +17.0% | [-79, +113] |
All (best-soft, unguarded): +21.1% [+6, +38] · n 592 — includes one-sided captures
NegBin sim, leak-aware; TOTAL match corners (audited). Canonical rows use the no-arb two-way guard: only lines where BOTH over+under were captured and the overround is a sane vigged book — kills mislabelled-longshot parse artifacts. Corrected 2026-07-08: leak-aware model (before_date = match date), capped non-bettable prices dropped (>21.0), fixture-clustered 95% CI (corners lines correlate within a match). Prices are TOTAL match corners (audited: not team/home/away corners). 1X2 is graded on its own panel below (Win Market recovered per-book from the preserved raw). No historical league game-stat odds on disk (odds_input/sgo are player-prop only) — leagues game-stat backtest not possible.
Goals (match O/U)
ACTUALWC · leak-aware · no-arb guard · fixture-clustered CI · n 185
| Cut | n | Win% | ROI | 95% CI |
|---|---|---|---|---|
| All | 185 | 50.3% | -15.3% | [-27, +2] |
| ● Value (edge ≥ 12.5%) | 30 | 50.0% | +9.3% | [-37, +62] |
| ● Best (edge ≥ 25%) | 4 | 75.0% | +66.7% | [-8, +145] |
Independent-Poisson on leak-aware total xG (Dixon-Coles tested 2026-07-09: no calibration gain — indep-Poisson retained).
Team goals
ACTUALWC · leak-aware · no-arb guard · fixture-clustered CI · n 288
| Cut | n | Win% | ROI | 95% CI |
|---|---|---|---|---|
| All | 288 | 59.4% | -16.1% | [-26, -3] |
| ● Value (edge ≥ 12.5%) | 33 | 39.4% | +9.2% | [-65, +144] |
| ● Best (edge ≥ 25%) | 6 | 16.7% | -68.2% | [-100, +15] |
Independent-Poisson on leak-aware team xG (Dixon-Coles tested 2026-07-09: no calibration gain — indep-Poisson retained).
Result (1X2)
ACTUALWC · leak-aware · no-arb guard · fixture-clustered CI · n 72
| Cut | n | Win% | ROI | 95% CI |
|---|---|---|---|---|
| All | 72 | 33.3% | -13.3% | [-33, +13] |
| ● Value (edge ≥ 12.5%) | 11 | 27.3% | +53.3% | [-100, +188] |
| ● Best (edge ≥ 25%) | 2 | 50.0% | +45.0% | — |
Backparsed from the preserved raw (Win Market per-book, run-up series; latest pre-KO snapshot graded). Model = leak-aware independent-Poisson xG (Dixon-Coles tested: no gain, retained). Settled on the 90-minute result. 3-way no-arb guard.
Cards (match O/U) · PILOT
ACTUALWC · leak-aware · no-arb guard · fixture-clustered CI · n 16
| Cut | n | Win% | ROI | 95% CI |
|---|---|---|---|---|
| All | 16 | 50.0% | -6.4% | — |
| ● Value (edge ≥ 12.5%) | 1 | 100.0% | +65.0% | — |
| ● Best (edge ≥ 25%) | 0 | — | — | — |
PILOT (capture expansion, QF onward). Referee-adjusted NegBin cards model, leak-aware; cards = SUM(yellow+red); WC referee fill is 0 so the ref term is inert here. PILOT — n too small to read; CIs are the point.
Held — evaluated, blocked (reported, not faked)
- cards_history: R32/R16 card prices are NOT recoverable: 0/353 preserved raw files carry any priced card/shot market (odds-store sweep 2026-07-08). Cards start at the QFs via the capture expansion (Stats Betting tab).
- shots_history: Same — Match Total Shots / SOT prices absent from all preserved raw; capturable from the QFs onward.
- player_sot_perbook: Player-prop per-book ladders (beyond goalscorer) not in the odds store — deferred (scope).
League capture (Aug 2026): armed, window-aware — 0 league fixtures in the 3-day window as of 2026-07-10T09:13Z. window-aware: 0 requests until league fixtures within 3d of KO