What the engine is — and what it is not
The BoxMidfield Engine is a structured decision framework. It does not simply pick the team with the shortest price, the highest league position or the best recent result. It asks what the available evidence says the match is most likely to look like, then identifies the selection that best expresses that match story.
The current production workflow is human-reviewed. The score is an evidence rating, not a mathematically guaranteed probability and not a claim that a result is certain. Missing or conflicting information can reduce the rating or stop a prospect from being published.
The evidence the engine investigates
Usually the latest 5–10 competitive matches, with home and away splits considered separately where they are relevant.
League position, opposition quality, goals for and against, and whether recent results are consistent with the underlying performance.
xG/xGA, shots, shots on target, possession and chance creation are used when a dependable source supplies them. Missing metrics are never invented.
Recent head-to-head evidence, styles, venue effects and matchup-specific strengths are supporting evidence rather than automatic rules.
Confirmed injuries, rotation, team news, congestion, cup context and other reliable information can materially change the assessment.
Coverage depth, conflicting sources and uncertainty matter. A strong-looking fixture with poor information quality can be downgraded or passed over.
How evidence becomes a prospect
The engine first forms one primary explanation of how the match is likely to unfold. Contradictory selections should not coexist unless credible new information changes the thesis.
The selection should capture the thesis without demanding more from the match than the evidence supports. Protection and robustness come before chasing a more aggressive outcome.
A prospect becomes stronger when several different evidence groups point in the same direction. One impressive statistic on its own is not enough.
Weak coverage, late team changes, contradictory form signals or unusual competition conditions reduce confidence rather than being ignored.
The final selection, match story and score are reviewed before publication. The reviewer can reject a technically qualifying fixture when the evidence is not robust enough.
Once published, the prospect is kept in the record and later settled as Won, Lost or Void. The result is used to evaluate the framework rather than being discarded.
How the BoxMidfield View score is formed
The score is a 0–10 evidence rating produced from the combined case, not a simple average of a fixed list and not a direct “90% chance” statement. The publishing template also carries a separate Estimated Probability used for Fair Price and EV calculations. There is currently no public claim that a 9.0 score equals a 90% outcome probability.
The final rating reflects the combined strength and robustness of the case. Estimated Probability is a separate model estimate; Implied Probability comes from the published price; Fair Price is derived from the model estimate; EV compares the two. BoxMidfield deliberately avoids publishing fake precision by pretending that each evidence group has a universal fixed weight across every league and competition.
What the tiers mean
The strongest evidence band in the current BoxMidfield scale.
A strong case with more uncertainty or less alignment than Open Net.
A published prospect with a weaker or less complete evidence case.
A higher tier means stronger evidence at publication time. It does not mean the prospect cannot lose.
Rules designed to protect the quality of the prospect
One primary match thesis per fixture unless credible new information materially changes it.
The engine favours the least demanding selection that accurately captures the evidence.
Credible in-play information such as a red card, injury or tactical change can supersede the pre-match assessment when clearly documented.
Late team news, rotation and lower-information competitions can create useful context, but lower data quality also increases uncertainty.
Different prospects should not quietly duplicate the same underlying match risk.
Results are tracked by rating band so future scores can be compared with what actually happened.
How BoxMidfield proves the process
Every published prospect should leave a record. The app displays settled outcomes and performance by tier, and the Stats page now shows rolling 7-day, 30-day and all-recorded performance. Winning and losing results remain part of the record.
Internally, the methodology also values calibration and closing-line comparison as research tools. These are used to test whether the framework is improving; they are not presented as guarantees to users.
What BoxMidfield will not do
It will not invent xG, injuries, lineups or other missing metrics. It will not call a score a certainty. It will not remove losing prospects to improve the historical record. And it will not publish a stronger-looking number simply because a user would prefer a more confident answer.