CourtsideIQ / Matchup Engine 2.4
The game.
Through our lens.
The matchup before tipoff. The evidence after it.
One
basketball job. A different way to read the game.
Historical results · Role context
Versioned forecasts ·
Measured error
Loading the game board…
01 / Game selection
On the schedule.
The model / ciq-matchup-2.4.0
Understand the forecast.
Then judge the result.
Evidence, not certainty.
Availability is game-specific, source-attributed and expires six hours after publication. Missing reports do not mean healthy. Verified outs receive zero minutes. Redistribution requires observed same-team workloads with and without the absent player; no replacement starter is invented. Questionable/doubtful reports show conditional scenarios without injury probabilities. Returns do not automatically restore a full workload.
Full baseline and rotation method
Minutes follow the last five same-team rotation observations, weighted toward the latest game. Explicit DNPs inform opportunity, not current injury status. Previous-season minutes are an opening-game fallback, not a season-long anchor. A team-specific nine-to-ten-player core prior distinguishes regular workloads from reserve opportunities. An eleventh regular requires at least three recent non-blowout games with meaningful usage. For core players with at least three recent appearances who played the latest game, a 35% blend toward their meaningful played workload reduces dilution by earlier DNPs and brief appearances. This does not confirm a lineup or guarantee that reserves will participate. A weighted 240-minute ceiling protects the established core before fringe players; missing minutes are never filled. Per-minute production shrinks toward a player's prior season (or the observed league) with a 300-minute prior. Recent form receives 25% weight. Estimated possessions adjust opportunity; a strongly shrunk opponent defensive rating makes a bounded points adjustment.
Early-season estimates are provisional. An injury or lineup source is not assumed. Rest and home court are context, not invented numerical bonuses. The role-based Matchup Advantage describes the opponent environment—it is not a betting probability.
Saved before the game. Measured after it.
Opponent vs. Role compares production by players with a role verified before their game. Per-possession normalization and an all-role team comparison isolate the incremental role environment. Small samples shrink toward neutral. Numerical effects stay off unless disjoint walk-forward testing supports a metric. No historical roles are backfilled from today's labels; no direct defensive assignments are implied. Original 2.0 and 2.1 forecasts remain unchanged. Read the separate OVR evaluation ↗
Only server-recorded forecasts saved before tipoff count toward performance. Recalculations create new versions. Historical games without an original run show “Pregame projection unavailable.” DNPs and missing actuals are not scored as zero production.
NBA.com source & actual archive coverage
Source: NBA.com. NBA.com usage terms apply. Missing imports are not proof that no games took place.