01
Projection
A rules-based estimate blends recent form, minutes, rest, and opponent allowance with an eligible model output when sufficient history exists.
Methodology
Projection is not history. History is not probability. Matchup context is not a promise. Contender keeps those boundaries visible from input through saved result.
PHMƒ
02
Observed history
Hit rates report wins, losses, pushes, and eligible sample counts for the selected side. They describe a window of past games, not a forecast.
03
Matchup
Opponent allowance, rank, trend, pace proxy, venue, and rest are descriptive context with definitions next to every value.
04
Probability
The selected-side probability is an uncalibrated model estimate conditional on the player participating. It is not a measured historical frequency.
05
Freshness
Every analysis records its source, observed time, eligible seasons, and latest eligible game date so the evidence can be audited later.
06
Limitations
Late scratches, role changes, unexpected minutes, injuries, and dependence between events can change outcomes after the snapshot is produced.
Data boundary
- 01
Observe
Schedules and completed box scores enter a versioned data snapshot.
- 02
Cut off
Only information eligible before the selected event is used for a historical run.
- 03
Compute
The engine records method labels, unavailable paths, and the latest eligible game date.
- 04
Preserve
A saved result retains its exact inputs, evidence, output, and timestamps.
Limits matter
Late scratches, role changes, unexpected minutes, injuries, and dependence between events can invalidate assumptions. Contender makes those limits part of the product, not fine print.