Skip to content

Rotation Models

Rotation models forecast playing-time allocation rather than player value. Their first target is a player's share of a team's total minutes in its next game. This includes whether a player appears and how much the player is used.

The initial family member is deliberately parameter-free:

Model Forecast Inputs
L0-MSP v0.0 Opening team-game minute share Prior final-game shares plus transaction-derived opening candidates
L20-MSP v0.0 Cumulative share across first 20 team games Prior-season total minutes plus transaction-derived opening candidates
L20-MSP v0.1 Cumulative share across first 20 team games L20-MSP plus a strictly preseason gap-returner prior
L20-NAIL-MSP v0.1 Cumulative share across first 20 team games L20-MSP plus completed prior-season NAIL-RAPM
L20-NAIL-MSP v0.2 Cumulative share across first 20 team games L20-MSP plus the exact target-season preseason NAIL forecast
L20-NAIL-MSP v0.3 Cumulative share across first 20 team games Source-aware role evidence for continuous players, gap returners, and cold starts
Forward Availability v0.1 Season medical-availability share Initial benchmark; superseded by v0.2
Forward Availability v0.2 Season medical-availability share Exposure-shrunk first state; clears age-only frozen benchmark
Forward Conditional Minutes v0.1 Preseason minutes per available game In development: aging and exposure-shrunk player state
AC-MSP v0.1 Game-level conditional minute allocation Observed availability mask; all regulation regular-season games
L1-MSP v0.0 Next team-game minute share That player's prior team-game minute share only
L5-MMP v0.1 Next team-game minute share Five prior team-game minute totals

NAIL-RAPM remains a rating model. Rotation models will eventually provide the player and five-man exposure forecasts needed to turn ratings into game and season predictions.