PER90My betsAccount

Value board

Bets where the model’s probability and the bookmaker’s price disagree by enough to matter. Every row that ever appears here is timestamped before kickoff, priced against a named book, and scored against the closing line on the public record. Nothing is deleted, including the losers.

Published edges

11

all time — every one is permanent

Prop prices captured

46,387

17 books quoting

Model

v20260812-1340

35,163 projections, latest run

Rejections logged

39438

every candidate that failed a gate

Live board

11 published calls on upcoming fixtures. Each was logged at the price shown and will be scored against the closing line whatever happens next. Prices last confirmed 21:12.

Stakes are quarter Kelly of the bankroll you typed, already capped per bet (2.5%) and across correlated bets (5% per match, 7.5% per referee — six edges under one strict referee are one opinion, not six). Your bankroll stays in this browser. This is a calculator you operate, not betting advice; the same figures are in the downloadable record as fractions.

How to read a row

The board shows, per bet: player and market, the line, the best available price and which book offers it, the model’s fair price, and the edge. Stake sizing lives in the bankroll calculator — a tool you operate, not advice printed next to a pick — and every advised figure is in the downloadable record.

Fair price
The decimal odds at which the bet would be break-even if the model's probability is right: 1 divided by that probability. A book price above fair is value; below it is the book's margin working on you.
Edge
How far the book's price sits above fair, as a percentage of expected value. The board publishes at a minimum of 4% — and demands more when the lineup is not confirmed: 7% on an expected lineup, 10% with none.
Stake
Quarter Kelly, capped at 2.5% of bankroll per bet — and capped again across correlated bets: all edges in one match share a 5% budget, all edges under one referee share 7.5%, split proportionally. Never full Kelly: full Kelly on a mis-specified model is ruinous, and every model is mis-specified. Sizing appears in the bankroll calculator, not beside each pick.
Confidence
Tracks lineup certainty, not conviction. "Provisional" means no lineup is out and minutes are estimated from start rates — those rows carry the highest edge threshold precisely because they are the least certain.

Publication rules

Every candidate bet passes all of these or it is not shown. Each rejection is logged with its reason — 39438 logged so far — so a quiet board is checkable, not just claimed.

GateRequirement
Minimum edge4% with a confirmed lineup · 7% on an expected lineup · 10% with none
Player sampleAt least 30 appearances with recorded minutes
Referee sampleAt least 15 matches officiated, else the league-mean referee effect is used and the edge bar rises by 2 points
Calibration marginEach market's measured worst calibration miss is added to its edge bar — a market whose deciles stray 7 points needs 7 more points of edge. A market whose calibration failed, or could not be demonstrated where the product bets, publishes nothing at all
Price freshnessBook price snapshotted within the last 30 minutes
Probability sanityModel probability between 10% and 90% — an extreme edge is usually a data error, not a gift
Cross-book sanityIf another book's implied probability differs by more than 25%, the line is suspected mismatched and rejected
Minutes floorProjected minutes of at least 20
Settlement matchThe book's settlement definition verified to match the stat the model prices
Exposure capsAdvised stakes are capped across correlated bets, not just per bet: all edges in one match share a 5%-of-bankroll budget and all edges under one referee share 7.5%, split proportionally — six fouls overs driven by one strict referee are one opinion, not six

Thresholds live in one config file on the model side, not scattered through code, and a change to any of them is recorded in the decisions log with the model version — so the record stays interpretable against the rules that produced it.