About Legwork

Multis, with the legwork done.

Where the prices come from

Accounts you connect yourself, and your key for each one stays on your device: sent with the request, used once, never stored.

• The Odds API: live odds from Australian bookmakers. Free 500 credits a month. Without it there is no pad.
• PuntersEdge: Australian and New Zealand racing, fourteen books on every runner. Free 1,500 credits a month. Without it there is no racing card.
• Betfair Exchange: the real market price, with no bookmaker margin in it. The free delayed key shows prices up to about three minutes old. It works from the phone only, because Betfair refuses requests from cloud servers.
• Squiggle: free AFL fixtures, results and a public model ensemble. No key needed.
• Cricsheet: cricket match data from Cricsheet, used under the Open Data Commons Attribution License 1.0. No key needed. It carries no prices: it is history the engine is tested against, not odds.
• Retrosheet: baseball line-ups and the weather at first pitch, used to test whether a late change of starting pitcher moves a price. No key needed, no prices in it. The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at 20 Sunset Rd., Mahwah, NJ 07430.

The board is fetched once a day, after 7 am, and everyone shares the same copy, so opening the app again costs nothing. A racing card is bought when you open the Racing tab or pull to refresh. Nothing polls, and a refresh shows the cost before it spends it: “About 5 credits for today's racing card. 467 of 500 left this month.”

The models, and what they say

Two strategies carry a model: Elo team ratings, and a Dixon-Coles goals model for football. Neither replaces the market's price; each corrects it, by a weight a regression actually fitted, and the correction is small. A model that has not shown it adds anything to the market's price prints no tickets at all and says why.

No model here beats the closing line yet. On the AFL, Elo scored a log loss of 0.609 against the market's 0.534. On the EPL, Dixon-Coles scored 1.002 against 0.932. The market is still the better forecaster.

Closing-line value is measured against the market's close, devigged and net of commission, with the book the bet was struck at left out of the vote. The figure carries its count and its interval, and the interval is the test: the whole of it has to clear zero before the app will say a strategy beats the close, and it has to fall below zero before NO EDGE is stamped. Anything straddling zero is “early”, however many legs it has.

A backtest replays past seasons a day at a time and reports the return, the strike rate, the drawdown and the closing-line value, with a range under the headline ROI. The racing history is the Betfair starting price and nothing else, so a racing run says what one book cannot test: there is no mispricing to find with a single column, and the starting price is the close, so there is no closing-line value either.

What it never does

• Never takes a bet, and never holds money.
• Never tells you to bet. Every ticket prints the chance it actually has, and the outrageous one prints a chance nobody would call a tip.
• Never claims more than it has measured.
• No same-game multis. Two legs from the same game are not unrelated, so the real chance of both is nothing like the two prices multiplied.
• No in-play. Everything on the pad is still to start.
• Nothing on a timer. Prices are bought when somebody asks for them.
• No analytics, no ads, no tracking.

Your data

What you record (placed tickets, stakes, results, settings) stays on your device, and deleting the app removes it.

Seeing the day's live board needs a sign-in while access is by invitation. The sample day at /demo needs nothing and saves nothing. See the privacy policy for what does and does not leave the device.

Publisher

Legwork · support