What "settled in public" means

Every tip site claims an accuracy figure. Almost none of those figures can be checked, because the site chooses which picks to count, when to start counting and what to do with the ones that lost. A record you cannot audit is marketing.

The ScoreSage verified results page is built the other way round. Every pick the models publish is written to a ledger before kick-off, settled automatically against the real result, and shown with its score, win or lose. Nothing is added afterwards and nothing is removed. This article is the rulebook behind that page, so that you can judge the record on its actual terms.

What goes into the record

The ledger tracks the picks published on the prediction boards:

Each pick is recorded with its market, the fixture, the probability or tier it was published at, and later its outcome and the final score.

When a pick is frozen

A pick is published before kick-off with its probability and tier, and it is not edited afterwards. If the model's view changes as team news arrives, the board can show a different pick tomorrow, but the one already published stays on the ledger exactly as it was. That is the difference between a record and a highlight reel.

How settlement works

Once a match finishes, the settler compares each pick with the final result and marks it won or lost. A few rules keep that honest:

What the headline number includes

The overall figure at the top of the results page combines first-half goal, BTTS, over 2.5 and under 2.5 picks. Two markets are handled differently on purpose:

Both rules are fixed in the settlement function, not decided pick by pick.

The numbers right now

As of 8 September 2026, all time:

Market Won Lost Hit rate Settled
Overall (headline) 2,441 1,076 69.4% 3,517
First-half goal 1,256 430 74.5% 1,686
Both teams to score 823 473 63.5% 1,296
Over 2.5 goals 227 105 68.4% 332
Under 2.5 goals 135 68 66.5% 203
First-half over 1.5 (excluded from headline) 466 618 43.0% 1,084
Corners (own row until earned) 8 5 61.5% 13

The live figures, with 7-day and 30-day windows, are on the results page. They will differ from this table by the time you read it; that is the point.

Why a hit rate alone can mislead

A hit rate says how often a pick lands. It says nothing about whether backing it made money, because that depends on the price. The break-even strike rate for any decimal price is simply one divided by the odds:

Decimal odds Strike rate needed to break even
1.33 75%
1.43 70%
1.54 65%
1.67 60%
1.82 55%
2.00 50%

A 74.5% first-half goal record is only profitable if the average price taken was above about 1.34; first-half goal lines in the covered leagues are often priced around 1.30 to 1.45, so the margin is thin and depends entirely on which picks you back. A 63.5% BTTS record breaks even at 1.58, and BTTS is frequently priced above that. The odds converter turns any price into the strike rate you need.

The one market where the ledger records the price is corners, because the Expected Corners model only publishes a pick at bettable odds. That row therefore shows a return on investment as well as a hit rate. For the others, the record tells you how well calibrated the models are; profit is a question of the prices you can actually get.

How to audit it yourself

If a number on the results page cannot be reproduced from the per-pick feed, that is a bug, and it will be treated as one.

What the record does not tell you

It does not tell you how a pick would have performed at the price you personally could have taken. It does not include picks from Scouts, which keep their own public records, or accumulators, which are settled inside the builder for the member who generated them. And it does not smooth over losing runs: the all-time best winning streak on the first-half goal market is 22, and the ledger also contains every losing streak that came between.

Read the record for what it is, a calibration check on the models, published in full. For how those models are built, read how AI football predictions work; for the under 2.5 side of the goals model in particular, read AI under 2.5 goals predictions explained.