Methodology

For each of today's basketball games with a verified total line, the model projects the combined score and compares it with the line. When the projection sits meaningfully below the line and several independent signals agree, the game is flagged as an analytical Under candidate.

Every output is a model estimate. Outcomes are uncertain and nothing here is a guarantee.

Five projections are blended, each only when it has enough data: (A) scoring-pace and efficiency baseline from points scored and allowed against the league norm, (B) recency-weighted form over the last 3–10 games, (C) home/away split model, (D) a learned totals model trained on stored results, (E) head-to-head and market-anchored reading.

Disagreement between projections widens the expected range and lowers confidence.

A transparent summary of eight components: Pace Suppression, Efficiency Suppression, Recent Under Trend, Matchup Fit, Lineup Impact, Schedule/Rest, Market Gap and Data Quality.

A component with no measured data is shown as unavailable and left out of the blend — it is never guessed.

A game qualifies only when: the model gap is at least 2.5 points below the line, the quality score is at least 62, the estimated Under chance is at least 54%, data quality is at least 55, each team has at least 4 recorded games, the line is fresh, and at least two independent signals (pace, efficiency, recent trend, matchup, rest) score 60+.

A streak of recent unders counts as one signal at most, so it can never qualify a game by itself. Admins can tighten these rules; they cannot loosen them without out-of-sample evidence.

Dean Oliver's Four Factors (effective field-goal %, turnover %, offensive-rebound % and free-throw rate) and possession-based ratings (ORtg/DRtg) are the standard basis of basketball efficiency modelling (Oliver, Basketball on Paper, 2004; Kubatko et al., 'A Starting Point for Analyzing Basketball Statistics', JQAS 2007).

The connected data feeds do not currently supply box-score or possession data, so these inputs are marked Data unavailable. The model uses points scored and allowed per game as its efficiency proxy until a box-score feed is connected.

Lines and prices: SportyBet Ghana (the source of truth for availability) and The Odds API as a second price source. Final scores: The Odds API scores feed. Team history is built from our own stored results.

Player injuries, lineups, travel and in-game live data are not supplied by these feeds and are shown as unavailable.

Every published analysis is locked at tip-off with its timestamp, source line, projection and confidence. After the game, the final total, result and projection error are recorded. Losing results are never hidden.

Model Health reports calibration, Brier score, MAE/RMSE and hit rate by confidence band, with sample sizes, and flags drift when recent error grows.