How it
works

1. 01

   Read the schedule

   Every game of the season, with its week, stage, live score and clock, comes from API-Sports. An odds feed only lists what it's currently pricing; the schedule feed covers the whole season.

   source · API-Sports
2. 02

   Read the market

   We take every sportsbook the feed carries for a game, remove each book’s margin from its own prices, and average the fair chances that are left. The number of books varies by game and is shown on each card as “N books.” Removing the margin book by book means a book that charges more can’t tilt the average toward the favorite. The margin comes off by the power method, which takes more of it from the longshot, where books load most of it. Books usually open a game 1–2 days before kickoff, so most of the season has no market price at any given moment.

   source · API-Sports odds
3. 03

   Remove the sportsbooks' cut

   A sportsbook prices a two-sided game at roughly 104–108% instead of 100%. The extra is its margin. We strip it out so both sides add up to exactly 100%. This is what turns a price into a probability.

   method · de-vig
4. 04

   Rate what the market hasn't priced

   A game with no odds yet still gets a number from our own power rating, built only from past results. It weights margin of victory so one blowout can't dominate, regresses last season toward average, and caps any game at 95%. It's marked est and is never blended into a market price.

   model · Elo power ratings
5. 05

   Lock the call

   Two hours before kickoff, the call is saved: the side, the probability, the number of books and the time. Until then the board's pick is provisional and can still move. Once saved, it is never overwritten. The final pre-game price is saved separately as the closing line, and the database refuses a closing price once the game has started.

   method · capture
6. 06

   Read the news (AI)

   Before NFL games, an AI model reads the latest team news from ESPN's public NFL news feed, above all injuries and who is actually playing, and writes a short news read. It's shown separately, never blended into the pick, and graded against the result like everything else.

   AI · news read
7. 07

   Watch the game and alert

   During the game, our live win-probability model reprices every pick as the score and clock change. An email goes out only when all four conditions in section III are true. One email per game, never repeated. An alert reports a position, not advice; the cash-out happens at your sportsbook, which we have no connection to.

   model · live win probability
8. 08

   Grade it in public

   Every finished game is scored against the final result. We publish hit rate, Brier score (how far each stated probability was from what happened; 0.250 is a coin flip, lower is better), closing-line value, and what a flat stake would have returned, all in units, not dollars. Preseason and thinly priced calls are reported separately, with the reason, and nothing is removed from the record.

   output · the record

One alert we actually sent, from the call to the final →

1. 01

   The position has to be in profit

   A bet returns the ratio between the price you took and the price now. At fair value, a ticket is worth the stake times the live chance divided by the chance we published, so it is in profit once our live model puts it above the probability we published before kickoff. No absolute level is required: a call made at 45% and now running at 55% is in profit, and one made at 85% and still at 85% is not, however commanding it looks. A floor of 59% is also applied.

   floor · 59%
2. 02

   It has to have improved on what we published

   At least 10 percentage points clear of the probability we gave it before kickoff, which keeps windows rare and each one substantial.

   margin · +10pp
3. 03

   The offer has to clear your stake

   A cash-out is the sportsbook buying your ticket back at the live price, then taking another cut, typically 5–15% under fair value. So fair-value profit is not enough: we only alert once an offer 10% under fair would still return more than you staked. Your stake size doesn't change this: $130 and $1,300 cash out at the same percentage. An early lead usually means the stake back plus a few percent, not the full win, which only exists if the pick wins.

   book's cut · 10%
4. 04

   The game has to still be running

   Cash-out only exists before the final whistle. Once a game ends the bet simply settles and there was never a decision to make, so an alert then would be an email about something you cannot act on.

   rule · in play only

1. 01

   Pull the odds

   Every market starts as real bookmaker prices. For each fixture we read the moneyline (home/draw/away) from every bookmaker that prices it, then average them into a consensus — more books, more reliable.

   source · The Odds API
2. 02

   Remove the vig

   Bookmakers price every market slightly above 100% — that overround is their margin. We strip it out (“de-vig”) so the probabilities are fair and sum to exactly 100%. This is the single most important step.

   method · de-vig
3. 03

   Fit a goal model

   For each fixture we solve for the pair of Poisson scoring rates whose implied win/draw/loss best matches the fair market. That score matrix gives us every goal market — over/under, BTTS, clean sheets, the most-likely scoreline and the full correct-score grid.

   model · Poisson
4. 04

   Render the desk

   The dashboard consumes normalized JSON from server-side routes, so the interface is identical whether the data is live or sample. Upstream calls are cached and shared across visitors to respect the free quota.

   output · the board

NFL, college football

API-Sports (schedule, scores, odds)

NBA, NHL, MLB, soccer

ESPN's public scoreboard (the DraftKings line, one book)

World Cup archive

API-Football and The Odds API

Golf

Data Golf (its own model, so not de-vigged)

News read

An AI model, reading ESPN's NFL news feed

Win / Draw / LossW/D/L

The chance the home team wins, the match is drawn, or the away team wins. The three add up to 100%, de-vigged from bookmaker moneyline odds. Used on the soccer board, where a draw is a result and can be the pick, and in the World Cup archive.

Expected goalsxG

The average number of goals the model expects each team to score in this match. It is the scoring rate that feeds every goal market below — not a prediction of the exact final score.

Both teams to scoreBTTS

The probability that both teams score at least one goal (i.e. the final score is not a clean sheet for either side).

Clean sheetCS

The probability a team concedes zero goals — i.e. keeps a clean sheet. 'CS BRA' is the chance Brazil's opponent fails to score.

Over / UnderO2.5

The probability the two teams combine for more than a given number of goals. 'O2.5' is the chance of 3 or more total goals; the Under is simply 100% minus the Over.

Most likely scorelineMost likely scoreline

The single exact score with the highest probability under the model. Even the likeliest scoreline is usually well under 20% — football is high-variance, so treat it as the modal guess, not a forecast.

Correct-score gridCorrect-score grid

A heatmap of every exact scoreline's probability (home goals down the side, away goals across the top). Brighter cells are more likely; it is the full picture the 'most likely scoreline' is drawn from.

De-vigged (vig removed)VIG

Bookmakers price every market slightly above 100% — that overround is their margin (the 'vig'). We strip it out so the probabilities are fair and sum to exactly 100%. On the game boards it comes off by the power method, which takes more from the longshot than the favorite, because that is where books load it. The 'VIG' figure is how much was removed.

Bookmaker consensusbooks

On the NFL and college boards these numbers are not one bookmaker's view — they are an average across every bookmaker that priced the market, each book's margin removed before averaging. 'N books' is how many fed this fixture; more books means a more reliable consensus. The NBA, NHL, MLB and soccer boards use a single book, the DraftKings line from ESPN's scoreboard, with its margin removed the same way.

Consensus odds (decimal)ODDS

The actual decimal price a bet would get, averaged across bookmakers — still including the margin. Decimal odds of 2.50 means a winning $1 stake returns $2.50 ($1.50 profit) — the same price a US book shows as +150. These are the real prices behind the de-vigged percentages.

Win probabilityWin %

Each team's chance of winning the game, from a de-vigged consensus of bookmaker moneylines. Football, basketball, hockey and baseball are two-way markets (no draw priced), so the two percentages add up to 100%; soccer adds the draw, see Win / Draw / Loss. The split bar underneath shows the same number visually — the center tick is 50/50.

Point spreadSpread

The bookmakers' points handicap for the game, shown against the home team. 'Chiefs -3.5' means the market makes the Chiefs a 3.5-point favorite: they must win by 4+ to 'cover'. A plus number means the home side is the underdog and can lose by up to that many points and still cover.

Cover probabilityCover %

The chance the home team beats the spread — wins by more than the handicap — de-vigged from the spread market's two prices. It hovers near 50% by design: bookmakers set the line so both sides are roughly a coin flip.

Game total (over/under)O/U

The market line for the two teams' combined points, priced as Over or Under. 'O 44.5 52%' means a 52% de-vigged chance the teams combine for 45+ points. Higher totals mean the market expects a shootout; lower, a defensive grind.

Model pickModel pick

The side the de-vigged bookmaker consensus makes the favorite, with its win probability. Because these probabilities come from the market itself, betting the pick at market prices has no built-in edge — it is a read on expectations, not betting advice.

Playoff seedSeed

The team's current playoff position in its conference: seeds 1–4 go to the four division winners (ordered by record), seeds 5–7 to the three best non-winners (wild cards). Seven teams per conference make the playoffs; the 1 seed gets the only first-round bye.

Point differential+/-

Season points scored minus points allowed. A positive number means the team has outscored its opponents overall — a steadier quality signal than the win-loss record, which can swing on close games.

Anytime touchdownAnytime TD

The probability a player scores a touchdown at any point in their next game, de-vigged from bookmaker anytime-TD props. Rushing and receiving TDs count; a quarterback's passing TDs do not.

Before kickoffBefore kickoff

The call was written to the database while the game was still unplayed, and the timestamp says when. It matters because the alternative is pricing a finished game from odds that outlive it: honest, but reconstructed, and you would have only our word that the number was not chosen once the score was known. Only a captured call can be graded without that caveat.

Brier scoreBrier

The average squared error of the stated probability. Lower is better, and 0.250 is what you would score by calling everything a coin flip. It exists because a hit rate alone flatters a 51% call that wins and punishes a 90% call that loses, when the first was barely an opinion and the second was a real claim.

Closing-line valueCLV

How the board's price compares with where the market finished before kickoff. The closing line is the sharpest price a market produces, so beating it consistently is the usual evidence that a model knows something. Measured in percentage points, and over a small sample it is mostly noise.

Closing lineClosing line

The last price the market showed before kickoff. It is the price the most money has been bet into, which is why it is treated as the market's best guess and used as the benchmark rather than the opening price.

Fair priceFair

The decimal odds a probability implies with no margin attached — 50% is 2.00, 80% is 1.25. A bet struck at the fair price has an expected value of exactly zero by construction, so a profit at fair prices is variance rather than edge. A real bookmaker prices below fair; that difference is how they make money.

Unitu

One flat stake. Results are quoted in units rather than currency because this site has never taken a bet and a dollar sign would imply that it had. Ten units a call means every call is staked the same, so no single result can dominate the record.

Return on investmentROI

Profit divided by everything staked, not by the number of bets. Backing ten calls at ten units each stakes 100 units, so a 12-unit profit is a 12% ROI. It is the figure that makes a long run of small stakes comparable with a short run of large ones.

Hit rateAccuracy

The share of calls that named the winner. Read it beside the Brier score rather than alone: a board that only ever calls heavy favorites will post a high hit rate while saying almost nothing, and one that takes real positions will post a lower one while being better informed.

CoverageCoverage

How much of the finished slate could be graded at all. A game the odds feed never quoted carries no call, so nothing was ever claimed about it and it counts as neither a hit nor a miss. Coverage sits beside the record because a board can always look more accurate by saying less.

No priceNo price

No bookmaker in the feed was quoting this game when the board last looked, so there is no market to de-vig and the board makes no call. It is ordinary rather than broken — the feed carries a minority of the college slate, and most lines only appear a day or two before the game — and a game with no price never enters the record.

Cash-out windowCash-out window

A published call that is now worth more in play than the price it was published at, while the game is still running. That is the only period in which a sportsbook's cash-out is worth taking: once the whistle goes the bet simply settles and there was never a decision to make. The board reports the window; the exit happens at your own sportsbook. Alerts cover the NFL, college football and the NBA so far.

Pre-game emailPre-game email

One email before the first game of a slate, listing every pick on the record for it, sent for the NBA, NHL, MLB and soccer to members who follow that sport. It is not an alert: it goes out before anything has happened and makes no cash-out claim. Choose your sports on your account page.

Variance, not edgeVariance

A run of winning calls priced from de-vigged odds is not evidence of beating the market. The fair price already contains everything the market knows, so betting it breaks even in the long run and any short-run profit is the swing you would expect from chance. The record here measures whether the probabilities are well calibrated, not whether following them makes money.