Hindcast — score the forecasts you already made

Score the forecasts you already made.

Calibration, closing line value with the bookmaker's margin taken out properly, and an honest interval around your results. No tips, no selections, no view on tomorrow. Free, unlimited, in your browser. Nothing is uploaded and nothing is stored.

Hindcast never tells you what to predict. It only tells you how well you have predicted — including when the honest answer is that your record cannot yet tell.

Reliability diagram — the hero of the product

A worked example, 280 forecasts

Stated probability on the horizontal, how often it actually happened on the vertical. On the 45-degree line you meant what you said; above it you were underconfident, below it you were overconfident. This example is overconfident above 60% and slightly underconfident below 30% — which is most people.

Door A

You bet against a price. Did you beat it after the margin came out?

Score a betting record

Door B

You forecast against a resolution date. Were your probabilities worth anything?

Score a forecasting record

I tested this on my own three samples before I let anyone else near it. Measured the way almost everybody measures it, my closing line value came out at +6.1%, +3.5% and +0.8%, and I felt quite good about that for roughly a day. Then I removed the bookmaker's margin the correct way — normalising across every outcome in the market, rather than deflating the single price I happened to take. The same three samples came back at +2.1%, −0.3% and −3.0%. Two of the three had lost to the close. This is not a quirk of my record. Raw CLV goes positive for every outcome in the market when a book widens before the event, so a whole population of bettors can each hold a spreadsheet showing they beat the close while collectively beating nothing. If you have been tracking CLV and you have not been de-vigging across the full market, you have been reading the bookmaker's margin and calling it your edge. It takes about forty seconds to find out which.

What it cannot do, before you upload anything

It cannot tell you whether you will win next month. It cannot tell you why. It cannot rescue a small sample and it will not pretend otherwise — most people who run this find they cannot yet demonstrate skill, and that is the expected result rather than a failure. It cannot see the forecasts you did not log, it cannot know when a forecast was actually made, and it cannot tell you whether an edge is bankable.

Read what this cannot tell you

Free means free

The verdict is free, unlimited, permanent and accountless. Paid tiers exist to publish and to print — not to unlock the answer.

See what costs money, and what never will