Our AI performance

Every football prediction NT Apex publishes goes into this record before the match is played, and stays in it once the result is in. Two rates come out of that archive: Bankers counted by themselves, and every prediction taken together.

The charts, the monthly table and the download files are three views of the same rows, so a match cannot be counted in one and missing from another. Nothing is added to the record after a result is known.

The first settled day is May 19, 2021 and the most recent is September 17, 2026, which puts 1,872 days of football in between. Each of those days is one line in the archive, with its counts beside it.

The table on this page shows the last 12 months. The file behind the download button starts at the first day and keeps the bad runs in with everything else.

Bankers sit at the top of the confidence scale, the few matches a day the model rates nine or better out of ten. Counted over 2026, that is 1,494 picks with 1,080 winners, a rate of 72.3%.

Overall is the full output: 276,793 settled predictions, 184,363 of them correct, 66.6%. A strict selection should land more often than everything the model says, and putting the two side by side is how you can see whether it does. One blended figure would hide it.

Correct tips divided by settled tips. Each rated match carries one main tip, published before kickoff and settled on the final score. A fixture that is postponed or abandoned never reaches a final score, so it drops out of the count instead of going down as a loss.

The headline rates weigh matches, not days, so a full Saturday card counts for more than three midweek fixtures. A tip is right or it is wrong: nothing is part-won and nothing is voided.

Once a day, after the previous day's matches have finished and been settled. That is why the last complete day here is September 17, 2026 rather than today, whose fixtures are still open.

Every panel on the page reads the same daily file, so they cannot drift apart. A day is written once, with the result it had, and it is not re-graded or quietly tidied up afterwards.

The CSV holds the daily record in raw form: one row per day with the date, how many Bankers settled that day and what share of them came in, then the same pair of columns for all predictions. Every chart above can be rebuilt from those five columns.

The spreadsheet goes a level deeper, one row per match with the competition, both teams, the tip we published and how the match ended. Neither asks for an account. To spot-check the last few days without downloading anything, All Matches shows the same fixtures beside their results.

Because a day is a tiny sample. On a quiet day the model rates only four matches highly enough to be Bankers, and four picks can land on five values: none, one, two, three or all four. A late equaliser moves the whole day.

A rate gets steadier the more matches it covers, which is why the monthly table and the all-time figures are printed here too, each one next to the sample it came from. Read the month before you read the day.

All of them, more than 700 across a year: top divisions in Europe and beyond, second and third tiers, domestic cups and two-legged continental ties. A title race and a midweek third-division fixture carry the same weight in these totals.

Nothing is filtered. No country is dropped, no competition is set aside and no month is left out because it went badly.

Download the two files and count again. Everything printed on this page comes out of those rows, so any figure here can be checked against the source instead of taken on trust.

You can also go back to the pages the picks were published on: All Matches keeps recent fixtures beside their results, and the day's strongest picks stay on Bet of the Day after kickoff. Losing tips are never deleted. They stay in the daily, monthly and all-time counts alongside the winners.