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

A calibrated sports predictor that publishes its own scorecard against a sharp market -- losses, nulls, and retractions included.

Every number below is copied verbatim from a committed artifact and wears its receipt: source path, as-of date, and verification status. No edge is claimed anywhere.

POSITIONING calibration, not profit · edge_claimed: false on every artifact
The measured sharpness result -- MLB in-game
+0.7327 CRPS

Model total-runs forecast is sharper than the market's at end of inning 7 (n=55, 95% CI [+0.4131, +1.0810]). MODEL_SHARPER_PROVISIONAL -- still provisional on a small sample. A sharpness result measured leak-free, not an edge.

And the honest reading
match, do not beat

Across in-game win-prob checkpoints the market is as sharp or sharper. Model Brier 0.2377 vs market 0.2067 on MLB (n=78,986); NBA end-Q1 favors the market too. Both readings are true. We publish both.

Cross-sport calibration scoreboard

model vs market -- paired Brier / CRPS delta, 95% CI, verdict

as of 2026-07-22T20:15:22.041778+00:00
sportmarketcheckpointnmodel − market95% CIverdict
nbawinprob (Brier)end_q11592
delta -0.0084 CI [-0.016101811879082918, -0.0007999346168341686] MARKET_SHARPER_PROVISIONAL-0.0084
[-0.0161, -0.0008]MARKET SHARPER
nbawinprob (Brier)halftime1593
delta -0.004 CI [-0.009833149837413686, 0.0015363951420276217] UNDERPOWERED-0.0040
[-0.0098, +0.0015]UNDERPOWERED
nbawinprob (Brier)end_q31593
delta 0.0011 CI [-0.0027845115004708075, 0.005201263747489016] UNDERPOWERED+0.0011
[-0.0028, +0.0052]UNDERPOWERED
nbawinprob (Brier)q4_under51593
delta 0.0019 CI [-0.00098643160153798, 0.004781044260357816] UNDERPOWERED+0.0019
[-0.0010, +0.0048]UNDERPOWERED
mlb_ingameCRPShome_margin|end_inning_320
delta 0.4499 CI [-0.020569384058188415, 1.0687464621667424] UNDERPOWERED+0.4499
[-0.0206, +1.0687]UNDERPOWERED
mlb_ingameCRPShome_margin|end_inning_526
delta 0.8555 CI [0.32594001284687374, 1.488773471523803] UNDERPOWERED+0.8555
[+0.3259, +1.4888]UNDERPOWERED
mlb_ingameCRPShome_margin|end_inning_622
delta 0.9583 CI [0.2408788621981401, 1.8046151457494153] UNDERPOWERED+0.9583
[+0.2409, +1.8046]UNDERPOWERED
mlb_ingameCRPShome_margin|end_inning_721
delta 1.066 CI [0.2656424426666572, 1.9601757169298957] UNDERPOWERED+1.0660
[+0.2656, +1.9602]UNDERPOWERED
mlb_ingameCRPShome_margin|end_inning_821
delta 1.794 CI [0.8720949949082061, 3.0589789365143907] UNDERPOWERED+1.7940
[+0.8721, +3.0590]UNDERPOWERED
mlb_ingameCRPStotal_runs|end_inning_349
delta 0.0891 CI [-0.3031783173062472, 0.5063543617299131] UNDERPOWERED+0.0891
[-0.3032, +0.5064]UNDERPOWERED
mlb_ingameCRPStotal_runs|end_inning_554
delta 0.2604 CI [-0.052614231788668975, 0.5751412231831872] UNDERPOWERED+0.2604
[-0.0526, +0.5751]UNDERPOWERED
mlb_ingameCRPStotal_runs|end_inning_654
delta 0.6392 CI [0.31739428723201235, 0.9632194949072426] MODEL_SHARPER_PROVISIONAL+0.6392
[+0.3174, +0.9632]MODEL SHARPER
mlb_ingameCRPStotal_runs|end_inning_755
delta 0.7327 CI [0.4130762214262852, 1.0809591352514807] MODEL_SHARPER_PROVISIONAL+0.7327
[+0.4131, +1.0810]MODEL SHARPER
mlb_ingameCRPStotal_runs|end_inning_849
delta 1.4201 CI [0.9658843906709079, 1.8976610619874428] MODEL_SHARPER_PROVISIONAL+1.4201
[+0.9659, +1.8977]MODEL SHARPER
mlb_pregameCRPS (total_runs)pregame300
delta -0.0523 CI [-0.1657, 0.0482] UNDERPOWERED-0.0523
[-0.1657, +0.0482]UNDERPOWERED
soccer_intlwinprob (Brier)home_win_prob|minute_6022
delta -0.0931 CI [-0.15696017645448831, -0.03359206672737812] UNDERPOWERED-0.0931
[-0.1570, -0.0336]UNDERPOWERED
soccer_intlwinprob (Brier)home_win_prob|minute_7517
delta -0.1075 CI [-0.1885062749116145, -0.03163929014699815] UNDERPOWERED-0.1075
[-0.1885, -0.0316]UNDERPOWERED
17 rows total. Losses and underpowered rows are shown, not filtered.open the full receipts ledger ->

nulls (351) outnumber confirms (168) 2.1x -- we publish our nulls

168
confirmed signals
351
null results
740
not testable
2.1x
nulls per confirm

Two ways in

Research labs

What you will not find here. No ROI, profit, or betting-edge claim. This is a calibrated predictor measured against the devigged close, not a track record. Several early headline numbers were retracted after this project's own instruments refuted them -- that story is a feature, on the retraction page, not a footnote.
Built by Neel Shah. Every figure traces to a committed JSON under scripts/platformkit/analytics_showcase/out/. Reproduce the whole set with python scripts/platformkit/analytics_showcase/check_all.py (52/52 on a bare clone). Truth source: docs/JOB_EVIDENCE_PACKET.md.