Demo -- read-only snapshot of the live system as of 2026-07-16T01:45:31Z; no live data, paper units only.
Skip to main content

Lab

Instruments & counterfactuals

Descriptive instruments and counterfactual replays. edge_claimed:false throughout -- these surfaces measure, they do not claim.

Outcome-free descriptive instruments over the joined in-game corpora. They track whether the two forecasters agree more and whether predictive uncertainty collapses as a game resolves. Entropy decay late in a game is expected as probabilities move toward 0 or 1; it is not a finding. Every panel is a textbook measure applied to our corpus, not a new method, and claims no edge.

Gap + entropy over game-time (MLB)

STANDARD_INSTRUMENTedge_claimed:false
.00.25.50.75.00t=1: 0.070t=2: 0.099t=3: 0.108t=4: 0.120t=5: 0.132t=6: 0.149t=7: 0.170t=8: 0.199t=9: 0.244t=10: 0.279MEAN GAP |MODEL-MARKET|.00.25.50.75.00t=1: 0.967t=2: 0.926t=3: 0.873t=4: 0.842t=5: 0.792t=6: 0.763t=7: 0.722t=8: 0.698t=9: 0.724t=10: 0.876MARKET ENTROPY (BITS)

GAME-TIME CHECKPOINT (1 to 10)

Mean |model-market| probability gap (amber) and mean market Bernoulli entropy in bits (blue) per inning checkpoint, MLB. Source: market_convergence.json.

Gap + entropy over game-time (soccer)

STANDARD_INSTRUMENTedge_claimed:false
.00.25.50.75.00t=0: 0.155t=5: 0.163t=10: 0.177t=15: 0.159t=20: 0.161t=25: 0.159t=30: 0.145t=35: 0.139t=40: 0.153t=45: 0.156t=50: 0.115t=55: 0.105t=60: 0.104t=65: 0.114t=70: 0.119t=75: 0.144t=80: 0.146t=85: 0.148t=90: 0.106MEAN GAP |MODEL-MARKET|.00.25.50.75.00t=0: 0.821t=5: 0.840t=10: 0.836t=15: 0.808t=20: 0.828t=25: 0.813t=30: 0.779t=35: 0.756t=40: 0.741t=45: 0.744t=50: 0.652t=55: 0.616t=60: 0.628t=65: 0.664t=70: 0.674t=75: 0.738t=80: 0.696t=85: 0.592t=90: 0.439MARKET ENTROPY (BITS)

GAME-TIME CHECKPOINT (0 to 90)

Same instrument on international soccer, per 5-minute checkpoint. Market entropy falls from 0.82 to 0.44 bits as matches resolve. Source: market_convergence.json.

Market-disagreement profile (MLB)

STANDARD_INSTRUMENTedge_claimed:false
gap bucketnmodel briermarket briermodel closer
<.02124150.1850.1840.449
.02-.05139340.2030.2030.459
.05-.10192350.2190.2170.408
>=.10334020.2830.2100.377
Rows bucketed by |model-market| prob gap; per-bucket Brier and the share where our model lands closer to the outcome. Where the two forecasters disagree most (>=.10), the market is sharper (model_closer_rate 0.38) -- shown, not hidden. Source: market_disagreement_profile.json.

Other staged instruments

STANDARD_INSTRUMENTedge_claimed:false
  • tick_microstructuremarket-microstructure MEASUREMENT (latency/cadence) -- NOT a trading signal
  • info_arrival_curvenot staged on this clone
  • residual_autocorrelationlag-1 sample autocorrelation of within-game signed residual (prob_t - terminal outcome), per (game_id, side) series ordered by ts
  • micro_absorptionpregame per-series consecutive-snapshot |devigged_prob| movement, bucketed by minutes-to-commence, from own scraped line_history
  • micro_closing_decayConsensus (median) devigged market P(home win) bucketed by hours-to-start into T-24h/T-6h/T-1h/close, scored Brier + log-loss vs the settled ESPN outcome; paired within-game close-vs-T-24h delta.
Each links to its raw artifact; a VALIDATION_PENDING chip means the file was not staged on this clone.