Transparency

Our public, verifiable track record. Every claim on this site must be reproducible here.

Model vs market

25%1 of 4 correct

10 leaguesNot enough settled matches yetEarly sample — 4 settled so far

Early sample — 4 settled predictions. These figures are provisional and not yet statistically meaningful.

Model RPS—lower is better

Market RPS—lower is better

RPS delta—not enough settled data yet

Graded vs closing line0settled predictions with a logged closing line — the rows the RPS pair is scored on

Mean CLV—

Positive CLV—

Settled predictions4

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 of the closing market line — beating it outright is not the claim.

Market-anchored vs model-only

Since the market anchor (ADR 0079) the served 1X2 probability is the fair market consensus plus a bounded lean from the model, so an anchored call can only sit a few points from the closing line by construction. The pooled figures above mix those calls with the ones the model made on its own. Here they are apart — each graded only on its own rows, with its own verdict.

Market-anchored

No settled calls in this group yet

Model-only

—RPS delta

4 settled · 0 graded vs the closing line

Not proven — too few rows to test

Record by market

The 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: settled count, model RPS, and — only where a real closing line was stored — the market's RPS, the gap and the bootstrap verdict. A market with no stored closing line is shown as unbenchmarked; we never invent a line to grade it against.

Settled predictions per market with model RPS, closing-line RPS where a closing line was stored, the gap between them and the bootstrap verdict
Market	Settled	Model RPS	Graded	Market RPS	RPS delta	Verdict
1X2headline	4	—	0	—	—	Not proven — too few rows to test

The AI's recent calls

What it predicted and how each one landed — the calls behind the number above.

The AI's most recent settled 1X2 calls: the predicted outcome, the result, and whether the call was correct.
Match	Date	Our call	Result
Arouca Estrela	11 Oct 2026	Home58%	1–0Home
Racing Santander Valencia	11 Oct 2026	Home47%	2–2Draw
Estac Troyes Marseille	11 Oct 2026	Home46%	0–2Away
Cagliari Juventus	11 Oct 2026	Away40%	1–1Draw

See all 4 calls

Public AI journalSimulated money

Follow the AI’s bankroll

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…

Historical AI bankroll replay

Historical simulation

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. This reconstruction is separate from the forward journal above; its balance is not carried into the new account.

Replay ending balance

$10,000

Started with $10,000

No picks have settled at a recorded price yet

The bankroll backs every pick this site publishes, priced at a bookmaker line snapped before the forecast. It sits at its starting capital until published picks settle with a real pre-kickoff price on record — never priced from an invented line.

Where all 4 settled predictions went:

No bookmaker odds recorded for the match at all: 4

Showing the latest 0 of 0 reconstructed bets. Download the complete history below.

Download complete history (CSV)Download complete history (JSON)

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+.

Hypothetical bankroll

Backtest

No value bets recorded yet

This ledger only stakes a pick when the model's price beats the best available bookmaker price by the value threshold. No qualifying edge means no bets — an empty book is the honest state, not an error.

4 settled predictions considered · 4 excluded for no recorded price.

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. Past performance does not guarantee future results. 18+.

Model vs market closing RPS by month (lower is better).
Month	Model	Market (closing)
Oct 2026	—	—

Signal scorecard — vs the closing line

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.

Beat the close50%

Mean signal CLV-3.3%

Settled signals4

Part of that number is the market's margin, not timing: on the 4 graded signals whose closing board stored a booksum, a bet struck at the market's own margin would have scored -5.6% on its own. Net of that, the signals scored +2.3% — 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.

Against the null

Our signals

-3.3%· 50% beat the close · 4 graded

Alerting on everything

—· — beat the close · 0 graded

Not proven either way — one side is still below the sample floor.

Null replayed over 14 Jul 2026 – 12 Oct 2026; the headline covers the full record. Both numbers carry the same margin bias, so only the gap between them says anything.

Arb · 1 settled · 100% beat close · +4.9% mean CLV
Model gap · 4 settled · 50% beat close · -3.3% mean CLV

Recent settled signals

The most recent settled signals with the price at detection, the fair closing price and each signal's closing-line verdict
Signal	Match	Outcome	Price at signal	Closing fair	Vs closing	Detected
Model gap	Aarhus vs Sonderjyske	Away	5.25Bet365	6.01	-12.7%Missed close	Mon 12 Oct, 06:05
Model gap	Aarhus vs Sonderjyske	Draw	4.61Pinnacle	4.88	-5.6%Missed close	Mon 12 Oct, 06:05
Arb	Atalanta vs Venezia	Away	6.29Pinnacle	6.00	+4.9%Beat close	Mon 12 Oct, 05:35
Model gap	Atalanta vs Venezia	Away	6.29Pinnacle	6.00	+4.9%Beat close	Mon 12 Oct, 05:35
Model gap	Atalanta vs Venezia	Draw	4.75Betano	4.74	+0.1%Beat close	Mon 12 Oct, 05:35

See the live signal feed

Why the headline anchors to the market

The headline 1X2 probability now anchors to the fair (vig-free) market consensus, with the model allowed a bounded lean of at most ±3 pp per outcome. This table is why: the same settled predictions scored as the old model headline, as the market consensus at the same instant, and as the anchored blend — plus the closing line as the end-state reference. We publish it even where the market wins; that honesty is the point.

Not enough settled predictions with a point-in-time market snapshot yet — the comparison fills in as matches settle.

lower is better

Audit chain

Every prediction is hashed into a Merkle tree and sealed on a schedule into a hash-linked block signed with Ed25519, so a forecast can be shown to have existed at seal time and not to have changed since. This is the current head of that chain; every row in the archive links to its own receipt, with the checks the server re-runs on request.

Height: 22; 3,496 predictions in this block
Block hash: d89e71ca1cc3eec6…; Sealed 12 Oct 2026
Signed: Yes — Ed25519; Public key: f5a303336353…
Timestamp anchor: Not yet anchored to an external timestamp

Stop-or-pivot triggers

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.

The three MASTER_SPEC §10 triggers with their current status, the measured value against its threshold and the sample size behind it
Trigger	Status	Value	Threshold	Sample
1X2 RPS vs closing line	Unprovable — sample too small	—	gap within +0.50 pp of the closing line, judged after 90 live days	0 graded of 4 settled
Signal CLV	Unprovable — sample too small	— all-time · — last 30 d	mean below zero with under half beating the close, in both windows	0 · 0 settled signals (min 100)
Free → paid conversion	Unmeasured — too few signups	—	rate at least 1% on 100 or more signups (90 d)	0 of 0 signups

Evaluated 12 Oct 2026