Our public track record. The aim is that every claim on this site can be checked here — where that isn't possible yet, we say so next to the number.
Expected from our own probabilities: 276 of 518 (53%), likely range 255–298
Bookmakers' favourite on the 514 matches with a stored price: 307 (60%); we got 308
A hit rate can only show whether we keep up with the bookmakers; the real test is beating their closing odds.
These figures count only the competitions in the public record. Results counts every competition we cover, so its numbers can differ.
Market RPS and CLV cannot be checked yet: they rest on closing lines and logged prices that the archive does not show per row.
Ranked Probability Score (RPS) — a measure of how close our probabilities were to reality, in plain terms: lower is better, zero would be perfect. We aim to land within ±0.005 RPS of the closing market line — beating it outright is not the claim. The RPS delta above is in the same unit.
We do not claim to beat the bookmakers. We publish calibrated probabilities and let the numbers stand.
Since the market anchor (ADR 0079) the served 1X2 probability is the market's vig-free price plus a bounded lean from the model, so an anchored call sits, by construction, no more than a few points from the market's price when the call was made. None of the calls the model made on its own has a logged closing line, so the RPS figures, “Graded vs closing line” and CLV above come only from market-anchored calls. Only the hit rate and the settled count mix the two groups. Here they are apart — each graded only on its own rows, with its own verdict.
514 settled · 513 graded vs the closing line
Does not beat the market4 settled · 0 graded vs the closing line
No closing line logged — cannot be gradedThe headline above is the 1X2 record. The model also predicts both-teams-to-score and over/under 2.5 goals, so here is every market on the same terms. Both RPS columns are scored only on the graded rows — the ones with a stored closing line (“Graded”) — so model and market are always compared on the same matches; “Settled” counts every settled prediction. A market with no stored closing line is shown as unbenchmarked; we never invent a line to grade it against.
| Market | Settled | Our forecast RPS | Graded | Market RPS | RPS delta | Verdict |
|---|---|---|---|---|---|---|
| 1X2headlinemarket-anchored | 518 | 0.1984 | 513 | 0.1983 | +0.0001 | Does not beat the market |
| BTTS | 514 | 0.2446 | 513 | 0.2448 | -0.0002 | Does not beat the market |
| Market-anchored | 508 | 0.2452 | 507 | 0.2449 | +0.0003 | Does not beat the market |
| Model-only | 6 | 0.1931 | 6 | 0.2350 | — | Too few matches to judge (fewer than 30) |
| O/U 2.5 | 514 | 0.2322 | 513 | 0.2327 | -0.0004 | Does not beat the market |
| Market-anchored | 508 | 0.2332 | 507 | 0.2327 | +0.0005 | Does not beat the market |
| Model-only | 6 | 0.1499 | 6 | 0.2270 | — | Too few matches to judge (fewer than 30) |
RPS levels are not comparable across markets: BTTS and O/U 2.5 have two outcomes, 1X2 has three. Read each row against its own market.
The Market RPS column cannot be checked here yet: the closing lines behind it are not shown per row in the archive.
What it predicted and how each one landed — the calls behind the number above.
| Match | Date | Our call | Result | Called it? |
|---|---|---|---|---|
| Rio Ave Sc Braga | 08 Nov 2026 | Away66%market-anchored | 0–1Away | |
| Estrela Benfica | 08 Nov 2026 | Away85%market-anchored | 2–3Away | |
| Atletico Madrid Barcelona | 08 Nov 2026 | Away49%market-anchored | 1–2Away | |
| Lens Marseille | 08 Nov 2026 | Home45%market-anchored | 1–2Away | |
| Fiorentina Juventus | 08 Nov 2026 | Away47%market-anchored | 2–2Draw | |
| Utrecht Nec Nijmegen | 08 Nov 2026 | Home40%market-anchored | 2–3Away | |
| 1899 Hoffenheim Fc Schalke 04 | 08 Nov 2026 | Home59%market-anchored | 2–1Home | |
| Vitoria Sc Fc Porto | 08 Nov 2026 | Away66%market-anchored | 1–2Away |
Bookmaker odds are turned off. You can turn them back on under Display at the bottom of the page.
Every tracked AI pick, from publication to result. See the price, the risk and the outcome together — including picks we could not price.
Loading picks…
A retrospective replay of settled 1X2, over/under 2.5 and both-teams-to-score calls, using a notional $10,000 and flat 1% stakes. Prices must have been recorded at or before the prediction. That cannot be checked here yet: neither the table nor the export shows each price's time and bookmaker. This reconstruction is separate from the forward journal above; its balance is not carried into the new account.
Started with $10,000
Return on the money staked so far: -0.3% (95% range -4.4% to +3.7%).
Against the market's fair closing price, the bets were worth on average -2.4%.
When the bets were placed, the market's own prices expected -2.5% on the money staked.
| Match | Pick | Odds | Stake | P&L | Balance |
|---|---|---|---|---|---|
| Inter Milano W vs Manchester City W10 Nov 2026 | Over 2.5 | 1.44 | $82 | +$36 | $8,580 |
| Paris Saint Germain W vs Bayern Munich W10 Nov 2026 | Over 2.5 | 1.50 | $82 | +$41 | $8,544 |
| Roma W vs Servette Chênois W10 Nov 2026 | Over 2.5 | 1.33 | $82 | +$27 | $8,503 |
| Inter Milano W vs Manchester City W10 Nov 2026 | BTTS yes | 1.73 | $82 | +$60 | $8,476 |
| Paris Saint Germain W vs Bayern Munich W10 Nov 2026 | BTTS yes | 1.53 | $82 | +$44 | $8,416 |
| Roma W vs Servette Chênois W10 Nov 2026 | BTTS yes | 1.70 | $82 | +$57 | $8,373 |
| Inter Milano W vs Manchester City W10 Nov 2026 | Away | 1.38 | $82 | +$31 | $8,315 |
| Roma W vs Servette Chênois W10 Nov 2026 | Home | 1.22 | $82 | +$18 | $8,284 |
| Paris Saint Germain W vs Bayern Munich W10 Nov 2026 | Away | 1.67 | $82 | +$55 | $8,266 |
| Real Madrid W vs Paris FC W10 Nov 2026 | Over 2.5 | 1.50 | $82 | +$41 | $8,211 |
| Real Madrid W vs Paris FC W10 Nov 2026 | BTTS yes | 1.67 | $82 | -$82 | $8,170 |
| Real Madrid W vs Paris FC W10 Nov 2026 | Home | 1.44 | $82 | +$36 | $8,252 |
| Deportivo de La Coruña W vs Valencia W09 Nov 2026 | Under 2.5 | 1.95 | $81 | +$77 | $8,216 |
| Deportivo de La Coruña W vs Valencia W09 Nov 2026 | BTTS no | 1.95 | $81 | +$77 | $8,138 |
| Deportivo de La Coruña W vs Valencia W09 Nov 2026 | Home | 1.57 | $81 | -$81 | $8,061 |
| Atletico Madrid W vs FC Levante Badalona W09 Nov 2026 | Over 2.5 | 1.55 | $79 | +$43 | $8,142 |
| Atletico Madrid W vs FC Levante Badalona W09 Nov 2026 | BTTS no | 1.80 | $79 | +$63 | $8,099 |
| Atletico Madrid W vs FC Levante Badalona W09 Nov 2026 | Home | 1.22 | $79 | +$17 | $8,036 |
| FC Koln W vs RB Leipzig W09 Nov 2026 | Over 2.5 | 1.65 | $79 | +$51 | $8,019 |
| FC Koln W vs RB Leipzig W09 Nov 2026 | BTTS yes | 1.59 | $79 | +$46 | $7,968 |
| FC Koln W vs RB Leipzig W09 Nov 2026 | Home | 1.75 | $79 | +$59 | $7,922 |
| Rio Ave vs SC Braga08 Nov 2026 | Over 2.5 | 1.72 | $76 | -$76 | $7,863 |
| Estrela vs Benfica08 Nov 2026 | Over 2.5 | 1.38 | $76 | +$29 | $7,938 |
| Rio Ave vs SC Braga08 Nov 2026 | BTTS yes | 1.80 | $76 | -$76 | $7,910 |
| Estrela vs Benfica08 Nov 2026 | BTTS yes | 1.83 | $76 | +$63 | $7,985 |
A hindsight replay of the same settled picks under other staking rules. Not a forecast — the best row here is not a recommendation.
| Strategy | End balance | Growth | Max drawdown | Bets placed |
|---|---|---|---|---|
| Flat 0.5% | $9,513 | -4.9% | 24.8% | 1,824 |
| Flat 1%Replay rule | $8,580 | -14.2% | 45.2% | 1,824 |
| Flat 2% | $5,833 | -41.7% | 73.7% | 1,824 |
| Quarter-Kelly (5% cap) | $13,697 | +37.0% | 25.6% | 540 |
Quarter-Kelly stakes only when the model claims an edge at the recorded price — picks with no claimed edge are sat out, which is why its row can stay at the seed. That is the honest result, not a display bug.
The Kelly stakes rest on the edge the model itself claimed at the recorded price — that is, on where the model disagrees with the market, and the model does not beat the market (see the model scoreboard). On 12 Oct those claimed edges came from the then near-flat both-teams-to-score and over/under 2.5 models, which gave 'yes' and 'over' roughly the same probability in every match (since anchored to the market, ADR 0143), and that one day carries a large share of this row's return. So however many bets it has placed, this row is not evidence of an edge.
Where all 2194 settled predictions went:
Showing the latest 25 of 1824 reconstructed bets. Download the complete history below.
Historical simulation, reconstructed from stored predictions and prices. It does not show bets actually placed or a real betting-account balance. It applies results in fixture order, without modelling overlapping open stakes or actual settlement times; it is not necessarily what a follower with limited available cash could have earned. The separate forward journal reserves stakes and records later settlements. This replay does not establish an edge. Past results do not guarantee future returns. 18+.
It stakes the 1X2 outcome where the model sees value against the best price — not the published call. So these are not the same bets as in the AI bankroll, and the result is not that of the quarter-Kelly row in its staking lab.
Backtested and hypothetical — not betting advice, and not a real ledger. Each pick is priced at the best bookmaker line snapped before kickoff; picks with no recorded price are excluded. Each price's time and bookmaker are not published yet, so that cannot be checked. Past performance does not guarantee future results. 18+.
Counting only the 12 competitions in the public record.
| Month | Published (market-anchored) | Market (closing) | Sample |
|---|---|---|---|
| Oct 2026 | 0.2013 | 0.2013 | 330 graded of 335 settled |
| Nov 2026 | 0.1932 | 0.1928 | 183 graded of 183 settled |
The historical AI bankroll replay above is one number, and one number can hide what made it: a book that only profits on odds-on favourites reads exactly like one that only profits on longshots. These are the same 1824 settled bets, split by the price we took and by the market. Hypothetical, like every figure here.
| Price | Bets | Return | Won |
|---|---|---|---|
| Odds-on (under 2.00) | 1532 | -2.5% | 61% |
| Mid (2.00–3.50) | 292 | +11.2% | 49% |
| Market | Bets | Return | Won |
|---|---|---|---|
| 1X2 | 614 | +9.0% | 61% |
| BTTS | 609 | -6.4% | 57% |
| OU25 | 601 | -3.6% | 61% |
A fixed 1 unit per bet across 1824 settled bets, at the best screened price recorded at or before the forecast. Each price's time and bookmaker are not shown yet, so that cannot be checked here. A price band nobody bet in is left out rather than shown as a flat zero. 370 of 2194 settled predictions had no price at the forecast and are left out.
Every signal whose match has a logged closing line is scored once it kicks off: the best screened price at detection vs the de-vigged closing consensus. Two caveats we would rather state than bury — it mixes line-shopping with timing, and it divides a price that still carries the bookmaker's margin by a margin-free fair price, so it reads low before any timing skill is involved. If the edge isn't there, this number says so.
Part of that number is the market's margin, not timing: on the 3,712 graded signals whose closing board stored a booksum, a bet struck at the market's own margin would have scored roughly -4.9% on its own. Net of that, the signals scored roughly -1.3% — which still includes whatever taking the best price across books earned, so read it as an upper bound on timing, and the replayed null below as the real comparison. Both figures are approximations: here the margin counts the same for every outcome, while the fair close puts more of it on longshots.
Against the null
Our signals
-6.1%· 26% beat the close · 3,712 graded
Alerting on everything
-0.3%· 46% beat the close · 557 graded
Not a skill test: under the current rule a signal only fires when the best price already sits above the market's fair price, so beating this null is built in.
The test that counts is whether the line holds: signals were taken +4.3% above fair on average, and the line then moved -1.1% by the close (86% still beat it, 109 graded).
Null replayed over 18 Aug 2026 – 16 Nov 2026; the headline covers the full record. Both numbers carry the same margin bias, so only the gap between them says anything.
| Signal | Match | Outcome | Price at signal | Closing fair | Vs closing | Detected |
|---|---|---|---|---|---|---|
| Steam | Häcken W vs Chelsea W | Draw | 6.50Betfair | 10.75 | -39.5%Missed close | Mon 09 Nov, 17:51 |
| Steam | Häcken W vs Chelsea W | Home | 11.25Marathonbet | 22.89 | -50.9%Missed close | Mon 09 Nov, 17:51 |
| Model gap | Fiorentina vs Juventus | Away | 2.27Betano | 2.30 | -1.4%Missed close | Sun 08 Nov, 19:20 |
| Model gap | 1899 Hoffenheim vs FC Schalke 04 | Away | 5.70Betano | 5.77 | -1.1%Missed close | Sun 08 Nov, 18:10 |
| Model gap | Villarreal vs Getafe | Away | 7.00Betfair | 6.93 | +1.0%Beat close | Sun 08 Nov, 17:10 |
| Model gap | Cambuur vs PSV Eindhoven | Draw | 5.50Betano | 6.19 | -11.1%Missed close | Sun 08 Nov, 16:40 |
| Model gap | SC Freiburg vs Union Berlin | Away | 7.00Marathonbet | 7.90 | -11.4%Missed close | Sun 08 Nov, 16:10 |
| Model gap | Angers vs Nice | Home | 3.62Marathonbet | 3.75 | -3.5%Missed close | Sun 08 Nov, 15:50 |
| Model gap | AZ Alkmaar vs Groningen | Away | 6.20Betano | 6.79 | -8.7%Missed close | Sun 08 Nov, 15:20 |
| Model gap | Real Sociedad vs Rayo Vallecano | Away | 4.40Marathonbet | 4.66 | -5.5%Missed close | Sun 08 Nov, 14:50 |
| Model gap | Randers FC vs FC Nordsjaelland | Away | 1.78Betano | 1.83 | -2.6%Missed close | Sun 08 Nov, 14:40 |
| Model gap | 1. FC Köln vs Bayer Leverkusen | Home | 4.20Betfair | 4.60 | -8.8%Missed close | Sun 08 Nov, 14:10 |
| Model gap | Atletico Madrid vs Barcelona | Draw | 4.45Betano | 4.69 | -5.2%Missed close | Sun 08 Nov, 14:10 |
| Model gap | Atletico Madrid vs Barcelona | Home | 3.34Marathonbet | 3.81 | -12.4%Missed close | Sun 08 Nov, 14:10 |
| Model gap | Monza vs Atalanta | Away | 2.07Marathonbet | 2.06 | +0.3%Beat close | Sun 08 Nov, 14:00 |
| Model gap | Genoa vs AC Milan | Home | 4.33Betfair | 4.31 | +0.4%Beat close | Sun 08 Nov, 14:00 |
| Model gap | Ipswich vs Bournemouth | Home | 4.05Betano | 4.07 | -0.5%Missed close | Sun 08 Nov, 13:40 |
| Model gap | Crystal Palace vs Liverpool | Home | 5.50Betfair | 5.63 | -2.4%Missed close | Sun 08 Nov, 13:40 |
| Model gap | 1899 Hoffenheim vs FC Schalke 04 | Draw | 4.65Marathonbet | 4.81 | -3.4%Missed close | Sun 08 Nov, 12:40 |
| Model gap | Vitória SC vs FC Porto | Away | 1.60Betano | 1.57 | +2.1%Beat close | Sun 08 Nov, 12:10 |
When a match has a market price, the headline 1X2 probability is anchored to that vig-free price, and the model may lean at most ±3 pp per outcome. The price comes from the sharp bookmakers that quote the match — that can be a single one — or else from the median of all bookmakers. The rows are scored on the same live, market-anchored predictions: the model's own number before anchoring and the market price it was anchored to — both stored when the prediction was made — plus the closing line on those of the rows with a logged close. The live row is the anchored number we actually published. Every row carries its own n. We publish the table even where the market wins.
| Probability source | n | Brier | RPS |
|---|---|---|---|
| Model alone (before anchoring) | 514 | 0.1883 | 0.2042 |
| Market's vig-free price (same instant) | 514 | 0.1834 | 0.1984 |
| Anchored headline (simulated)No comparable rows yet: the simulation is built only from settled predictions the model made on its own with a market snapshot from the same instant. | — | — | — |
| Closing line | 513 | 0.1830 | 0.1983 |
| Anchored headline (live) | 514 | 0.1833 | 0.1983 |
Brier = mean over the three outcomes, Σ(p−o)²/3; 0 is perfect, 0.667 is worst. Lower is better for both Brier and RPS.
Every settled market-anchored prediction stores the model's own number and the market price it was anchored to. From those two we rebuild the published number at several lean caps and score it against the market price alone, on the same predictions. A negative difference means the lean helped. The rule was set in advance: once a month, a market whose lean scores worse than the market price beyond chance has its lean set to 0. This check changes no published number.
| Lean cap | RPS vs market price |
|---|---|
| 0 (market price only) | +0.00000 |
| ±1 pp | -0.00014 |
| ±2 pp | -0.00018 |
| ±3 pplive | -0.00005 |
| ±5 pp | +0.00029 |
| ±10 pp | +0.00171 |
Not decided yet: at the live ±3 pp the difference from the market price is within chance. About 1,702 more settled predictions would be needed to see a difference of 0.001.
220 settled predictions are left out: published without a market price, missing one of the two stored numbers, or blended with the market price rather than leaned from it.
| Lean cap | RPS vs market price |
|---|---|
| 0 (market price only) | +0.00000 |
| ±1 pp | +0.00011 |
| ±2 pp | +0.00035 |
| ±3 pplive | +0.00067 |
| ±5 pp | +0.00168 |
| ±10 pp | +0.00539 |
Not decided yet: at the live ±3 pp the difference from the market price is within chance. About 3,844 more settled predictions would be needed to see a difference of 0.001.
222 settled predictions are left out: published without a market price, missing one of the two stored numbers, or blended with the market price rather than leaned from it.
| Lean cap | RPS vs market price |
|---|---|
| 0 (market price only) | +0.00000 |
| ±1 pp | -0.00001 |
| ±2 pp | +0.00003 |
| ±3 pplive | +0.00012 |
| ±5 pp | +0.00089 |
| ±10 pp | +0.00243 |
Not decided yet: at the live ±3 pp the difference from the market price is within chance. About 3,861 more settled predictions would be needed to see a difference of 0.001.
222 settled predictions are left out: published without a market price, missing one of the two stored numbers, or blended with the market price rather than leaned from it.
Every prediction is hashed into a Merkle tree and sealed on a schedule into a hash-linked block signed with Ed25519. The chain shows that a prediction existed at seal time. The timestamps are self-attested until they are anchored externally. The last prediction before a match is made minutes before kickoff and is usually sealed only after kickoff, so the chain does not yet prove that a call existed before the match. This is the chain's latest block when the page loaded; every row in the archive links to its own receipt, with the checks the server re-runs on request.
MASTER_SPEC §10 names three conditions under which we change course: model RPS out of band against the closing line after three live months, signal CLV consistently negative, and free-to-paid conversion below the floor. We claim to follow these rules; this is where anyone can check. Evaluated daily by the spec_check job — a trigger never fires on an unprovable sample, and the numbers are shown even where there is no verdict yet.
| Trigger | Status | Value | Threshold | Sample |
|---|---|---|---|---|
| 1X2 RPS vs closing line | Not due yet | +0.0001 | gap within ±0.005 RPS of the closing line, judged after 90 live dayslive since 11 Oct 2026 (date of the first settled prediction) | 513 graded of 518 settled |
| Signal CLVAction: remove 'value' marketing; reposition as insight and entertainment | Fired | -6.1% all-time · -7.0% last 30 d | mean below zero with under half beating the close, in both windows | 3712 · 2730 settled signals (min 100) |
| Free → paid conversion | Unmeasured — too few signups | 0.0% | rate at least 1% on 100 or more signups (90 d) | 0 of 19 signups |
Evaluated 17 Nov 2026 · live since 11 Oct 2026