How our predictions have held up

Our price model is the first of our own predictors to earn a graded record. Across 1,805 price-change calls it read the direction of a player's price move — up, down or flat — correctly 47.9% of the time. A three-way call gives roughly 33.3% to a coin, so that is better than guessing, though not by a margin worth leaning on. Its average miss was $29.4K, which is a lot of money to be wrong by on a single trade. We show this because grading ourselves by the same rule we apply to everyone else is the entire point of this page — a better number would have been more comfortable to publish, not more honest.

Corrected 25 August 2026. Up to round 25 we published price calls for players who could not change price — fewer than three games that season, or not playing that round. There were 2,154 of them, and every one was scored against a movement of zero. They have been withdrawn from the graded record. Our hit rate over the calls that remain is 47.9% rather than 27.4%; our average miss is $29.4K rather than $26.2K. The second figure is worse because the withdrawn calls were easy to be nearly right about.

A price call counts as correct when it gets the direction right, with any move under $3,000 treated as flat. The average miss is the mean gap, in dollars, between the predicted change and the real one.

Each buy the engine recommends is judged on what the player actually scored over his next three played rounds, against the median of every player the engine could have named that week. Across 138 graded calls, 61.6% beat that median. Every one of those calls came from the nightly form-value-fixture blend — when the method changes, this figure starts again under the new method's name rather than quietly absorbing the old record.

A caution on the sample: 138 calls across rounds 19–24 of a single season, all under this year's Stats Perform adjudication. Six rounds is a window, not a settled record, and we publish it because the ledger is graded, not because the sample is large enough to lean on.

What this figure covers: the buy calls this engine publishes each night across every squad it reads, chosen by the nightly form-value-fixture blend. It does not cover the trade of the round on your front page, which is picked for your squad alone by a different method — that one keeps its own record, and it will appear here under its own name once enough of its calls have graded.

For each buy the engine recommends, regret is the SC points the best affordable same-position alternative averaged over the next three completed rounds, less the pick's — priced as at the time of the call. Zero means the pick was the best affordable buy. A call only grades once all three of those rounds have been played and settled, so a short or part-settled window waits rather than counting.

What this figure covers: the buy calls the engine publishes each night across every squad it reads, chosen by the nightly form-value-fixture blend — the same calls the record above describes. It does not cover the two-trade planner's combinations, or the trade of the round on your front page, both of which are different methods over different squads. The comparator differs too: regret is measured against the single best affordable alternative, where the record above is measured against the median of the field.

The planner's graded calls have not yet reached the 20 the commentators on this page are held to — 8 have graded so far, so the rate is withheld until the record clears the same bar.

What this figure covers: the buy calls the two-trade planner served to readers who opened the trade board, chosen by the two-trade optimiser. Each buy is graded on its own merits rather than as part of the combination it was shown in, so a two-trade plan counts here as its separate buys and not as one call.

The YouTube channels and podcasts we track, ranked by their trade-in record. A trade-in call counts as correct when the player bought beat the median of his position over his next three played rounds. Captain and trade-out calls are tracked too; the fuller per-call breakdown lives on the commentators page.

Hit rates cluster in the low-to-high twenties. Trade-in calls are hard: beating the positional median over three rounds is a genuine bar, and no tracked commentator clears it much more than three times in ten.

When a tipster flags a call as high confidence, it ought to land more often than their run-of-the-mill medium call. Across the tracked field it mostly does not: only 5 of 16 commentators with a large enough sample show their high-confidence trade-ins beating their medium ones. Shown only where both bands clear 100 graded calls; smaller samples say nothing reliable and are left out.