mtgarbbuy globally → sell AU · MTG + Pokémon

inference calibration

Per-model calibration metrics for the inference-rebuild. Read-only. Source: inference_models × arb_predictionsarb_prediction_outcomes.

active model registry

activeversiongametierfit attrain rowsWF CRPSWF log-scorecov 50%cov 90%
v1.1-magic-tier3-2026-05-16T0320Zmagic32026-05-16 03:2010,144
v1.1-pokemon-tier3-2026-05-16T0145Zpokemon32026-05-16 01:457,628
v1.1-pokemon-tier3-2026-05-16T0423Zpokemon32026-05-16 04:237,628
v1.1-magic-tier3-2026-05-16T0144Zmagic32026-05-16 01:4410,144

coverage + proper score

Tier 2 (clearing-prob): proper-score = Brier (lower is better). Tier 3 (log-price): proper-score = mean |log-residual| (lower is better); 50% coverage should be ≈ 0.50.

tiergamemodelN w/ outcomecov 50%proper score
3magicv1.1-magic-tier3-2026-05-16T0320Z22,03958.0%0.7178
3pokemonv1.1-pokemon-tier3-2026-05-16T0145Z2,00654.8%0.8307

tier-3 residuals by stratum

log(realised) − log(predicted p50) per (game × price-band). Median should be ≈ 0 if unbiased; mean-abs is a sharpness proxy.

gameprice bandnmedian residmean |resid|
magicA: <$101,697-0.5480.651
magicB: $10-5012,672-0.3780.589
magicC: $50-2005,905-0.2820.900
magicD: $200-1k1,765+1.0641.096
pokemonA: <$101,665-0.8230.829
pokemonB: $10-50336-0.6840.817
pokemonC: $50-2005+2.4122.267