Forward Availability

Forward Availability forecasts a player's medical availability share for a future season. It is intentionally separate from playing time: a player can be medically available but receive no minutes because they are outside the rotation.

The target is available games divided by known rostered player-games, not games played. Its training panel and binary label are defined in the player-game availability contract.

Current Contract

For target season \(t\), the model uses only completed earlier seasons. An exposure-weighted quadratic age curve provides the pooled baseline. Each player's deviation from that baseline is carried forward with persistence \(\rho\), a gap adjustment, and a small standardized lagged-workload term:

\[ \operatorname{logit}(p_{i,t}) = a(\operatorname{age}_{i,t}) + \rho^{g_{i,t}}\left(z_{i,t-1} + w\,\widetilde{m}_{i,t-1}\right). \]

Observed availability updates the state with a beta-binomial-style pseudo-game prior. The first observed season starts from the age baseline, rather than treating a raw \(0/82\) observation as certainty:

\[ \widehat p^{\mathrm{post}}_{i,t} = \frac{q\,p^{\mathrm{prior}}_{i,t} + A_{i,t}} {q + N_{i,t}}. \]

Here \(A\) is available games and \(N\) is known rostered player-games. The promoted output is paired with Forward Conditional Minutes to form expected season minutes:

\[ E[\mathrm{minutes}_{i,t}] = P(\mathrm{available}_{i,t}) \times E[\mathrm{minutes}_{i,t}\mid\mathrm{available}_{i,t}]. \]

Current Production State: v0.2

The promoted configuration is:

Parameter	Value
State persistence \(\rho\)	0.50
Later-state update strength \(q_{\mathrm{update}}\)	60 pseudo-games
Initial-state strength \(q_{\mathrm{init}}\)	15 pseudo-games
Lagged workload weight	-0.25

It was selected on completed 2020-21 through 2022-23 targets and evaluated on frozen 2023-24 through 2025-26 seasons.

Mean frozen metric	Age-only control	v0.2
Exposure-weighted Brier score	0.06133	0.05816
Binomial log loss	0.56145	0.55168
Player-share MAE	0.19984	0.19145
Player-share RMSE	0.25888	0.25338

Artifact: artifacts/rotation/forward_availability/forward-availability-20260909T233439Z-e341ed5.

Update History

v0.1: Initial State Filter

The first release established the forward-only age curve, filtered player state, pseudo-game update, and frozen evaluation contract. It improved player share MAE but lost to the age-only control on probability-sensitive metrics, which exposed overconfidence in first observed seasons. It was not promoted.

v0.2: First-State Calibration

v0.2 changed only the first-state update: a player's first observed season now starts from the age baseline with its own tuned pseudo-game strength. This fixes the failure mode where a missed rookie season produced an implausibly near-zero forecast. For example, Chet Holmgren's 0 / 82 rookie observation led v0.1 to forecast 0.3% availability for 2023-24; v0.2 forecast 47.3% before his return, with 100.0% realized availability.

The v0.2 result cleared every aggregate frozen metric and is the availability component used by Win Projections.