How it works

American games are settled on a points margin, so we rate them differently. Every team carries a strength from its recent margins and a power rating that moves with each result (more for an upset, less for a favourite running up the score), and from those we work out how many points one side should beat the other by and how many points the game should have in all.

Those numbers are then weighed against the market's own line, which gets most of the say: the closing spread is the sharpest number in sport, and we have no reason to pretend otherwise. Around the expected margin sits the real spread of past results, which knows that NFL games are decided by three and by seven far more often than any bell curve would say. From it come the moneyline, every spread and every total you see.

Checked on seasons the fit never saw, our NFL moneyline chances come out a shade sharper than the market's, while at the closing line the spreads and totals add little to it: they are shown with the ladder of alternative lines, and the rule for a tip, above, decides whether any of them is one. Every tip is graded against the closing line like the football.

The NHL is rated the way the American leagues are: every team carries a strength from its recent goal margins and a power rating that moves with each result, and from those we work out the expected margin and the expected goals in a game, weighed against the market's own line where there is one. Hockey adds a wrinkle of its own: about a quarter of games are level after sixty minutes and go to overtime or a shoot-out. So the moneyline, the puck line and the total count the overtime and the shoot-out's goal, as the bookmakers settle them, while the 60-minute result is a market of its own whose draw is the league's rate of overtime games, lower when one side is clearly stronger.

DraftKings' lines come from ESPN's scoreboard, and Pinnacle's and other bookmakers' prices are read from their own pages. A game is tipped only when it is likely enough and a bookmaker's price is better than the market-backed chance says it should be. In the table an overtime or shoot-out defeat keeps its point, as in the league's own standings.

Every player on the two tours carries a rating that moves with each match, and one for each surface: a clay-courter reads higher in May than in July. A new player's rating settles quickly and a veteran's barely moves. The gap between two players' ratings, blended across the surface they are about to play on, becomes our chance of each winning; the curve that turns a gap into a chance is fitted on ten seasons of results, with a steeper one for best-of-five matches, where the better player has longer to prove it.

Checked on the last two seasons, which the fit never saw, the chances hold: when we say 70 %, the player wins about seven times in ten. Where Pinnacle's or an exchange's price reaches us, a tennis match is tipped only when the player is likely enough to win and a softer bookmaker's price is better than what those sharpest books say the match is worth; a match without such a price shows our chance and the fair price it implies and leaves the comparison to you. A retirement or a walkover voids a bet on the match, which is also how the bookmakers settle it.

Every fighter in the UFC and the PFL carries a rating that moves with each fight, settling fast for a newcomer and barely for a veteran, and a record of how their fights end: how often they finish, how often they are finished, how often they hear the final bell. Our chance of each hand being raised comes from the gap between the two ratings together with what each fighter brings to the cage: their ages on the night (after the ratings, the biggest single factor; youth wins), their height, reach and stance, the time since each last fought and how often each has been put away, all weighed on ten years of cards. Over the last two seasons it called 66 % of winners where the ratings alone called 58 %. The two records, the weight class and the scheduled rounds give the chance the fight goes the distance; a finish is split between a knockout and a submission by the fighters' own habits and spread over the rounds the way the division's fights end, which is where the total rounds and the method of victory come from.

DraftKings prices the UFC's fights in a free feed, so a UFC card shows the bookmaker's chance beside ours. A fight is tipped only when the fighter is likely enough to win and a bookmaker's price is better than what the sharpest books, Pinnacle and the exchanges, say the fight is worth, or once our own chances have beaten the bookmakers' over a long enough run of priced fights, which we check again every week. Until then no fight is tipped on our view alone: on the fights priced so far our view is level with the bookmakers', not ahead of them, and we would rather show you that than pretend. The method and the rounds are shown with their fair prices and left for you to weigh. A draw or a no contest returns a stake on the winner, as the bookmakers settle it; the PFL has no prices in any free feed, so its fights carry our chances and fair prices only.

Boxing has no free feed of results, so it starts from the market. Every bout The Odds API lists is on the board, with the bookmakers' prices beside it where we have them; where Pinnacle or an exchange prices a bout, that price is our chance of each fighter, with a small share for the draw, and a bout is tipped only when the fighter is likely enough to win and a softer bookmaker's price is better than it. A bout no sharp book prices shows the ratings' view, which starts level for every fighter and learns from each result we record, tempered by whatever price there is. The distance, the method and the rounds are shown from the fight model's curves and never tipped.

Results come from Wikipedia: every notable boxer's page keeps a record table, filled in by its editors within a day of a televised bout, and we read it through Wikipedia's open interface (the text is CC BY-SA) for the winner, the way the bout ended, the round and the time. Two thirds of the fighters on a typical week's list have such a page; a bout between two who do not is settled by hand from the official card or, if we cannot, is taken off the record ten days on and any pick on it returns its stake. A draw loses a bet on either fighter unless you backed the draw; a no contest returns every stake.

Every side carries a rating within its own game: a nation's Test, one-day and T20 sides are rated apart, as they play apart, and each franchise league is rated on its own. The rating moves with every decided match, quickly for a new side and steadily for an old one; a draw or a tie is half a point each, and a washout moves nothing. The home side's edge is measured a format: large in Tests and one-day internationals, where touring sides struggle, smaller in T20, and nothing on neutral ground. The gap between two ratings becomes our chance through a curve fitted on twenty-five years of results, with a curve of its own for the internationals of each format, the T20 leagues and the Hundred. A Test has a third way out: the chance of a draw is read from the year, since draws have grown rarer with every decade, and the winner's chance is the rest shared between the two sides.

Checked on the last two seasons, which the fit never saw, the chances called 65 % of one-day internationals, 72 % of T20 internationals and 72 % of the Tests that reached a result; the franchise leagues are the hardest to call, 55 %, because their sides are built to be even, and there a rating alone is worth little more than the toss. Where Pinnacle's or an exchange's price reaches us, beside the prices of the bookmakers we read ourselves, a match is tipped only when the side is likely enough to win and a softer bookmaker's price is better than what those sharpest books say, or once our own chances have beaten the bookmakers' over a long enough run of priced matches, which we check again every week. A match without a sharp price shows our chance and the fair price it implies. A tie, a drawn Test or a washout returns a stake on the winner, as the bookmakers settle it; on a Test's three-way result the draw pays, and an abandoned Test returns the stake.

Handball is settled on a margin of a few goals in a game of sixty, so it is rated like the American leagues: each side's strength from its recent margins and a power rating that moves with each result, from which come the goals one side should beat the other by and how many goals the game should have in all. The draw is the share of past results that finished level after accounting for the expected margin, about one game in twelve.

The matches come from the European federation's own site: the Champions League and the European League, men's and women's. There are no free prices for them, so the board shows our chance of each result, every handicap and every total with the fair price each implies, and no tips until a price feed exists.

A baseball game turns on a run or two, and a season has 162 of them, so a team's power rating moves only a little with each result and its strength is read from its recent run margins. From the gap between the two sides and the home side's edge come the runs one should beat the other by and the runs the game should have in all; the moneyline, the run line at a run and a half either way and the total runs are read off that spread, extra innings included.

The games and DraftKings' lines on the games to come are ESPN's. Where a line is quoted it anchors our expectation, so for now a baseball chance is close to what the price says; there is no free history of baseball prices to measure the ratings against, and until a season of our own lines has been graded nothing in baseball is a tip. The probable starting pitchers are shown on each card: they matter more than any team number, and the bookmakers' lines already carry them.

Rugby union and rugby league are settled on a points margin wide enough for a handicap to be the main market, and rated like the American leagues: each side's strength from its recent margins, a power rating that moves with each result and a home advantage worth about four points. The draw keeps its small share, a couple of games in a hundred, so the result market is three-way.

The matches are ESPN's: the Premiership, the Top 14, the United Rugby Championship, Super Rugby Pacific, the Champions Cup, the Six Nations, the Rugby Championship, the World Cup, the international tests and the NRL. ESPN quotes no prices on rugby, so the board shows our chance of each result, the handicaps and the totals with the fair price each implies, and no tips until a price feed exists. League tables show four points for a win and two for a draw, without bonus points.

A Grand Prix is rated the way a golf tournament is: not two sides but a field. Every driver carries a strength read from their recent finishes, a win worth most, a retirement counting as last, the old races fading; on the morning of the race the grid slot counts beside it, because where a car starts is the single best guide to where it finishes. The race is then run twenty thousand times, each car with a surprise of its own, and the chances of winning, of a podium, of the top six and of the points are the shares of runs in which each happened.

The calendar, the grids and the classifications come from the Jolpica API, the free successor of the Ergast Formula 1 archive. The board for a race goes out on the morning of the race, with Saturday's grid in it; in the week before, the outlook shows the chances from form alone, refreshed every morning. There are no free prices for the races, so the chances come with the fair price each implies and no tips until a price feed exists.