Model Report Card
Every sport at a glance
- MLBLive recordincludes corrections through Nov 16
Measuring in public. No persistent price edge is claimed.
1851–1387 · 57.2% hit · +9.7% per play, units +303.45u · 3,138 of 3,238 at licensed books · as logged +327.57u
MLB card and full conclusion - WNBALive record
Measuring in public. No persistent price edge is claimed.
65–48 · 57.5% hit · +8.2% per play, units +8.56u · 105 of 113 at licensed books · as logged +6.37u
WNBA card and full conclusion - College BasketballBacktest
The one sport where our probability beat the closing price.
StatIQ ahead · Brier .24468 vs .24541
College Basketball card and full conclusion - NBABacktest
The market wins. We ship tracker + best price + projections, never +EV.
Market ahead · pooled Brier gap +0.0117 (z +29.2)
NBA card and full conclusion - NHLBacktest
The market wins every market. Tracker + best price only, never +EV.
Market ahead · pooled Brier .2407 vs .2289 (z +32.5)
NHL card and full conclusion - MMA (UFC)Backtest
The close beat us in all six years. Tracker pattern only, never a pick product.
Market ahead · log-loss .642 vs .593
MMA (UFC) card and full conclusion - College FootballBacktest (dated) · live 2026 record 119–102includes corrections through Nov 16
Never scored against a closing-price archive. Treat accordingly.
Not measured
College Football card and full conclusion - NFLForward test
Forward test of a selection rule: mixed so far.
Focused 202–169 · +0.6% · Broad 246–211 · −0.9%
NFL card and the forward test
4 of the 7 sports with a model study have been scored against a banked archive of real closing prices. 3 have not, and their cards say so rather than borrowing another sport’s verdict. Backtest = a model run over past seasons, evidence about a method and never a record. Live record = plays published before the event and graded in the open. Forward test = a rule written down before the games, measured on them at closing prices on a flat one unit; no bets placed.
How to read this page
- BACKTEST means a model was run over past seasons and scored against archived prices. It is evidence about a method. It is not money, not a record, and never a promise.
- LIVE means every graded play was published before the event started and graded afterwards, in the open.
- Rates need at least 20 graded decisions before we print them. Under that they show a dash, in both directions.
- Every conclusion here carries the date it was measured, and each figure is checked against its study before it is published. A number comes down if the study no longer supports it.
Conclusions curated 2026-08-06. Live records are read from the API at page load, never stored here. Some analyses are not published.
On this page
What worked, what failed, what is unproven
One dated line per sport. Where the closing price beats us, we say so here first.
- MLB Live record
MLB, as of 2026-10-14: no persistent price edge is claimed. Every segment edge we tested in 2026 WNBA and MLB props dissolved as the sample grew (price/edge result, recorded 2026-07-20). Our probability has not been scored against the closing price, so there is no accuracy result either way. What we stand on is the live graded record and live calibration below.
Live, computed at page load · Sample: the live graded record below · Scope: No closing-price study committed for MLB
- WNBA Live record
WNBA, as of 2026-10-14: no persistent price edge is claimed (the price/edge result shared with MLB, recorded 2026-07-20). Our probability has not been scored against the closing price, so there is no accuracy result either way. What we stand on is the live graded record and live calibration below. The gap between the probability we displayed and the hit rate delivered, and its direction, are computed from the graded rows at page load, never typed.
Evaluated 2026-08-05 · Sample: the live graded record below · Scope: No closing-price study committed for WNBA
- College Basketball Backtest
CBB: in a walk-forward backtest scored at real 2025-26 closing prices, our probability beat the no-vig close. Unproven going forward: there is no live CBB record yet, and a backtest is not a record.
Evaluated 2026-08-06 · Sample: 24,681 two-sided props scored for calibration · Scope: Real 2025-26 closes, one snapshot ~14 min pre-tip
- NBA Backtest
NBA: no edge established across our segment tests through the 2025-26 season. The closing price beat our probability in every market, in both seasons scored. Projections only, never picks.
Evaluated 2026-08-06 · Sample: 193,342 prices scored · Scope: Real closes, 2024-25 + 2025-26 incl. playoffs
- NHL Backtest
NHL: the closing price beat our probability in every market across two seasons, including the one market where a smaller sample had suggested otherwise. No edge; no NHL product has shipped.
Evaluated 2026-08-05 · Sample: 164,099 prices scored · Scope: Real closes, 2024-25 + 2025-26 incl. playoffs
- MMA (UFC) Backtest
MMA (UFC): our fight-winner model is calibrated and better than a coin, but the closing price beat it in all six years scored, 2021-2026. No edge; no MMA product.
Evaluated 2026-08-06 · Sample: 2,680 fights scored · Scope: Real closes, 2021-2026, six years walk-forward
- College Football Backtest
CFB: never scored against a closing-price archive, so whether the CFB market is efficient is unproven either way. Our newer projector beat the older one in backtest; that is a comparison of our own models, not a record.
Evaluated 2026-08-05 · Sample: 476 backtest picks (2024-25 test window) · Scope: Close-snapshot backtest prices · no closing-price archive scored
- NFL Forward test
NFL forward test: mixed so far. Focused 202–169 (+0.6% measured return); Broad 246–211 (−0.9%). (includes corrections through Nov 16)
Live, computed at page load · Sample: Focused 371 graded, 0 pending; Broad 457 graded, 0 pending · graded weeks 5–9 · Scope: a selection rule, measured at closing prices on a flat one unit, no bets placed
MLB
LIVE · measuring in public. Price/edge: no persistent edge claimed. Accuracy vs the close: not measured. MLB, as of 2026-10-14: no persistent price edge is claimed. Every segment edge we tested in 2026 WNBA and MLB props dissolved as the sample grew (price/edge result, recorded 2026-07-20). Our probability has not been scored against the closing price, so there is no accuracy result either way. What we stand on is the live graded record and live calibration below.
- Scored against
- No closing-price study committed for MLB
- Sample
- the live graded record below
- Measured
- live, computed at page load
How our probability was scored against the closing price
- Our probability vs the closing price
- Not measured
Current evaluation, as of 2026-10-14
- Price/edge result
- No persistent edge. Every apparent segment edge we tested dissolved when the sample grew, so we claim none. Measured 2026-07-20 · Population: 2026 WNBA and MLB player props on our own board; each apparent edge was first seen on fewer than 150 plays, and the largest re-test covered a full season of 302 games · Source: Claim 4 on How we measure.
- Accuracy result (our probability vs the closing price)
- Not measured. No closing-price study has been committed for this sport, so we make no claim either way about whether our probability or the closing price is the better one. Never measured · Source: this card.
Earlier wording, dated
- 2026-07-20 to 2026-10-14: How we measure called the price/edge result “the prop markets we have measured are efficient”, while this card said “no efficiency verdict published” about the accuracy comparison: one word for two different results. Both were renamed on 2026-10-14; neither finding changed.
The graded record: every play published before the event Live record
| Era | W–L | Hit rate | Return per play | Units | Sample |
|---|---|---|---|---|---|
| Current model since 2026-07-28 | 1851–1387includes corrections through Nov 16 | 57.2% | +9.7% as logged +10.12% | +303.45u 3,138 of 3,238 at licensed books · as logged +327.57u | 3238 decisions |
| Previous model since 2026-07-12 | 359–322includes corrections through Nov 16 | 52.7% | −3.2% as logged −0.89% | −13.02u 412 of 681 at licensed books · as logged −6.08u | 681 decisions |
| Whole official record since 2026-07-12 | 2210–1709includes corrections through Nov 16 | 56.4% | +8.2% as logged +8.20% | +290.43u 3,550 of 3,919 at licensed books · as logged +321.49u | 3919 decisions |
Return per play is flat 1-unit staking on the price we locked, before any bonus, boost or fee. It is what happened, not a forecast. A model cutover starts a new era; the previous era stays published and unedited.
Source: the live MLB board · read at page load · graded through 2026-10-28 · same selection cutoffs as the Record tab, which rebuilds the same picks from the durable record ledger on its own grading sweep, so the two can sit several days apart: the same record at two freshness points, each stamped with its own graded-through date.
What we claimed vs what happened: current model era, 3238 graded decisions Live record
Calibration is shown to members: it needs the model's pre-game chance on each of the 3238 graded decisions. The wins and losses are in the table above.
Source: the live MLB board · computed at page load from the graded rows of the current model era
WNBA
LIVE · measuring in public; fewer than 20 graded decisions in the current model era, so no calibration verdict is printed yet. WNBA, as of 2026-10-14: no persistent price edge is claimed (the price/edge result shared with MLB, recorded 2026-07-20). Our probability has not been scored against the closing price, so there is no accuracy result either way. What we stand on is the live graded record and live calibration below. The gap between the probability we displayed and the hit rate delivered, and its direction, are computed from the graded rows at page load, never typed. Evaluated 2026-08-05.
The gap between the hit probability we displayed and what the record delivered is computed live from the same rows the record is graded on: fewer than 20 graded decisions in the current model era, so no calibration verdict is printed yet.
- Scored against
- No closing-price study committed for WNBA
- Sample
- the live graded record below
- Measured
- 2026-08-05
How our probability was scored against the closing price
- Our probability vs the closing price
- Not measured
Current evaluation, as of 2026-10-14
- Price/edge result
- No persistent edge. Every apparent segment edge we tested dissolved when the sample grew, so we claim none. Measured 2026-07-20 · Population: 2026 WNBA and MLB player props on our own board; each apparent edge was first seen on fewer than 150 plays, and the largest re-test covered a full season of 302 games · Source: Claim 4 on How we measure.
- Accuracy result (our probability vs the closing price)
- Not measured. No closing-price study has been committed for this sport, so we make no claim either way about whether our probability or the closing price is the better one. Never measured · Source: this card.
Earlier wording, dated
- 2026-07-20 to 2026-10-14: How we measure called the price/edge result “the prop markets we have measured are efficient”, while this card said “no efficiency verdict published” about the accuracy comparison: one word for two different results. Both were renamed on 2026-10-14; neither finding changed.
The graded record: every play published before the event Live record
| Era | W–L | Hit rate | Return per play | Units | Sample |
|---|---|---|---|---|---|
| Record since the restart since the Sep 2026 restart | 65–48 | 57.5% | +8.2% as logged +5.64% | +8.56u 105 of 113 at licensed books · as logged +6.37u | 113 decisions |
Return per play is flat 1-unit staking on the price we locked, before any bonus, boost or fee. It is what happened, not a forecast. A model cutover starts a new era; the previous era stays published and unedited.
This is the live record since the September 2026 restart only. The earlier window was wiped from the live board by a publishing bug and recovered pick by pick from our daily-board archive. It is kept separate and never added in; see it, auditable, on the record page.
Source: the live WNBA board · read at page load · graded through 2026-10-31 · same selection cutoffs as the Record tab, which rebuilds the same picks from the durable record ledger on its own grading sweep, so the two can sit several days apart: the same record at two freshness points, each stamped with its own graded-through date.
What we claimed vs what happened: current model era, 113 graded decisions Live record
Calibration is shown to members: it needs the model's pre-game chance on each of the 113 graded decisions. The wins and losses are in the table above.
Source: the live WNBA board · computed at page load from the graded rows of the current model era
College Basketball
BACKTEST · the one sport where our probability beat the closing price. CBB: in a walk-forward backtest scored at real 2025-26 closing prices, our probability beat the no-vig close. Unproven going forward: there is no live CBB record yet, and a backtest is not a record. Evaluated 2026-08-06.
- Scored against
- Real 2025-26 closes, one snapshot ~14 min pre-tip
- Sample
- 24,681 two-sided props scored for calibration
- Measured
- 2026-08-06
How our probability was scored against the closing price
- Our probability vs the closing price
- StatIQ ahead · Brier .24468 vs .24541
No forward record. No live CBB record exists. The season starts in November 2026; everything above is a walk-forward backtest scored at archived closing prices.
NBA
EFFICIENT · the market wins. We ship tracker + best price + projections, never +EV. NBA: no edge established across our segment tests through the 2025-26 season. The closing price beat our probability in every market, in both seasons scored. Projections only, never picks. Evaluated 2026-08-06.
- Scored against
- Real closes, 2024-25 + 2025-26 incl. playoffs
- Sample
- 193,342 prices scored
- Measured
- 2026-08-06
How our probability was scored against the closing price
- Our probability vs the closing price
- Market ahead · pooled Brier gap +0.0117 (z +29.2)
No forward record. No live NBA record exists, and none is kept: on StatIQ, NBA is projections only, never picks. Every number on this card is a walk-forward backtest scored against archived closing prices.
NHL
EFFICIENT · the market wins every market. Tracker + best price only, never +EV. NHL: the closing price beat our probability in every market across two seasons, including the one market where a smaller sample had suggested otherwise. No edge; no NHL product has shipped. Evaluated 2026-08-05.
- Scored against
- Real closes, 2024-25 + 2025-26 incl. playoffs
- Sample
- 164,099 prices scored
- Measured
- 2026-08-05
How our probability was scored against the closing price
- Our probability vs the closing price
- Market ahead · pooled Brier .2407 vs .2289 (z +32.5)
No forward record. No live NHL record exists, and no NHL product has shipped. Everything on this card is a walk-forward backtest scored against archived closing prices.
MMA (UFC)
EFFICIENT · the close beat us in all six years. Tracker pattern only, never a pick product. MMA (UFC): our fight-winner model is calibrated and better than a coin, but the closing price beat it in all six years scored, 2021-2026. No edge; no MMA product. Evaluated 2026-08-06.
- Scored against
- Real closes, 2021-2026, six years walk-forward
- Sample
- 2,680 fights scored
- Measured
- 2026-08-06
How our probability was scored against the closing price
- Our probability vs the closing price
- Market ahead · log-loss .642 vs .593
No forward record. No live MMA record exists, and no MMA product has shipped. Everything on this card is a walk-forward backtest scored against archived closing prices.
College Football
BACKTEST · never scored against a closing-price archive. Treat accordingly. CFB: never scored against a closing-price archive, so whether the CFB market is efficient is unproven either way. Our newer projector beat the older one in backtest; that is a comparison of our own models, not a record. Evaluated 2026-08-05.
- Scored against
- Close-snapshot backtest prices · no closing-price archive scored
- Sample
- 476 backtest picks (2024-25 test window)
- Measured
- 2026-08-05
How our probability was scored against the closing price
- Our probability vs the closing price
- Not measured
Current status, read from the live board at page load: the 2026 CFB board is live: 119–102 on 221 graded plays, every one logged before kickoff and graded in the open on the CFB board. That is the live record; the backtest above is a separate, dated evaluation of the model and is not part of it.
includes corrections through Nov 16CFB projection accuracy: ours against the closing line
Computed over 1,970 graded college football player-prop projections for games from Aug 29, 2026 to Nov 15, 2026: every one with our projection, its closing line and the final result. These are projections, not picks. Each row is the same prop, measured against the same result.
The closing line has been closer so far. The closing line was closer in all 3 markets. On average our projections missed by 36.90 yards, the closing line by 34.57 yards.
| Market | Closer | Props | Ours | Line |
|---|---|---|---|---|
| Receiving Yds | Line | 1,038 | 28.58 | 27.17 |
| Rushing Yds | Line | 621 | 34.88 | 32.69 |
| Passing Yds | Line | 311 | 68.70 | 62.99 |
Metric: average miss, the gap between the number and the final stat, in that stat’s own unit (a smaller number is closer); the “Ours” and “Line” columns are each side’s average miss. The closing line is the last line we stored before the game started. “Level” means the 95% interval of the per-prop difference in miss (ours minus the other side’s) includes zero, so the gap is within what chance produces at this sample size; the figures are shown either way. Nothing is excluded for how it came out: a row is left out only when it is not graded or lacks a projection, a line or a result.
NFL
Forward test NFL’s evidence is a forward test of a selection rule: the rule was written down before the 2026 season and is measured on games played since, at closing prices on a flat one unit, with no bets placed. Two overlapping tests, Broad and Focused; the Focused rows are also in Broad.
NFL forward test: mixed so far. Focused 202–169 (+0.6% measured return); Broad 246–211 (−0.9%).
- Focused (our selections): 202–169 · 54.4% hit · +0.6% measured return (+2.22 units) · 371 graded, 0 pendingincludes corrections through Nov 16
- Broad (control): 246–211 · 53.8% hit · −0.9% measured return (−4.32 units) · 457 graded, 0 pendingincludes corrections through Nov 16
Projection accuracy: the NFL comparison of our projections with the closing line, and with each player’s season average, is on the NFL forward test page, over the graded 2026 board rows it names. The record above is the forward test only.
The rule’s written commitment, dated before the season, is on the NFL 2026 commitment; every row is on the NFL forward test.