A prediction engine can produce a number for any fixture you feed it. Whether that number deserves publishing is a separate question, and answering it honestly is most of what separates a research tool from a tipping service.
CompactStats declines to pick on a substantial share of the games it covers. Each rule below was bought with evidence, and several of them cost us calls that looked good.
Audited across 1,150 historical games: our calls landed 51.4%, a coin flip. The structural cause is that a cross-border tie inherits an anchor from two unrelated domestic leagues. Full audit. We publish the percentage and make no pick.
If either side has fewer than three league games in the current season, its rating is mostly league average wearing a team's name. The model has no team-specific information to offer, so it offers none.
The site publishes top-20 boards of the most over-heavy and under-heavy teams in our data. When a fixture's call contradicts our own published board — an under lean involving a team we list as one of the most over-heavy sides — we found those calls historically ran about three points optimistic. Rather than quietly ignoring the contradiction, the row is marked and no pick is made.
The most expensive rule to adopt, because the calls we deleted were winning 75% of the time. Why we did it anyway.
Fixtures involving clubs from divisions we do not cover carry an explicit "limited data" badge and are excluded from our strongest-selection panels entirely.
Silence is a prediction too — it says "we do not know", which for roughly a third of world football on any given day is the accurate answer. The temptation to fill every cell of a table is strong, because empty cells look like a product with gaps. They are the opposite: they are the cells where a number would have been decoration.
Method: walk-forward testing — every prediction is made using only information available before kick-off, then graded against what happened. Studies run June-July 2026 on the site's own dataset. Questions: contact us.