DARKO (Daily Adjusted and Regressed Kalman Optimized) DPM is a Bayesian player impact metric developed by Kostya Medvedovsky and published via The Analyst. It uses a Kalman filter that treats each player's true talent as a time-varying quantity, updated daily as new game data arrives. The model incorporates a strong prior from box score data that decays as on-court sample size grows.
The Kalman filter begins with a box score-derived prior for each player and updates it with new plus-minus observations each game, with recent games weighted more heavily than older ones. The result is a daily estimate of each player's per-100-possession impact.
Converges to a reliable estimate faster than static-window RAPM, because the time-series approach extracts more information from early-season data. Methodologically distinct from EPM and LEBRON — not just another regression variant, but a fundamentally different mathematical approach. Excellent for in-season evaluation.
Forward-looking by design. DARKO's creator has explicitly stated that DARKO "cannot meaningfully be used in backwards-looking, MVP-type debates." The Kalman filter is optimized to predict future performance, which means it pulls dominant single-season outlier performances back toward prior expectations faster than EPM or LEBRON would. In historically exceptional seasons — 2015-16 Curry, 2021-22 Jokić — DARKO will be more conservative than other estimators.
DARKO DPM is scored in APEX's Overall Impact pillar at approximately 20% within-pillar weight (~6.6% of composite score). It is weighted below EPM and LEBRON specifically because of its forward-looking design, which makes it less suited to retrospective season evaluation. However, DARKO's faster early-season convergence is a genuine asset that justifies its inclusion: blending it with EPM and LEBRON means APEX's early-season rankings are more stable than they would be using EPM alone. See Known Limitations §8 in the APEX methodology for full discussion.