Overview

Six-agent pipeline: data → features → meta-scientist → eval → critic → report. Metrics update after every run.

Best log loss
0.6008
random = 0.6931
Experiments
92
across 6+ runs
Predictions logged
38
CLV ledger
Paper trades
20
¼-Kelly sizing
Paper PnL
$8540
settled trades
Avg CLV
vs closing line

Champion model

CHAMPION tabpfn — TabPFN - trimmed 10-feature core — log loss 0.6008
Critic: approved_with_caution: Audit pipeline broke down at every verification step, so promotion cannot be confirmed as genuine progress: 1. Red-flag audit failed entirely (no result.json for 84a05669) — unable to check for leakage, fold instability, or suspicious feature names. Its log_loss of 0.6008 sits right at the theoretical NBA moneyline ceiling (~0.60), which is plausible but leaves zero margin — would normally warran · promoted 2026-07-10

Convergence — best log loss over experiments

0.5870.6170.6470.6770.707random (0.693)#1 2026-04-13: 0.6374#2 2026-04-13: 0.6374#3 2026-04-13: 0.6374#4 2026-04-13: 0.6306#5 2026-04-13: 0.6306#6 2026-04-13: 0.6306#7 2026-04-13: 0.6261#8 2026-04-13: 0.6261#9 2026-04-13: 0.6176#10 2026-04-13: 0.6176#11 2026-04-13: 0.6176#12 2026-04-13: 0.6176#13 2026-04-13: 0.6175#14 2026-04-13: 0.6175#15 2026-04-13: 0.6175#16 2026-04-13: 0.6174#17 2026-04-13: 0.6174#18 2026-04-13: 0.6174#19 2026-04-13: 0.6174#20 2026-04-13: 0.6174#21 2026-04-13: 0.6174#22 2026-04-13: 0.6174#23 2026-04-13: 0.6174#24 2026-04-13: 0.6174#25 2026-04-13: 0.6174#26 2026-04-13: 0.6174#27 2026-04-13: 0.6174#28 2026-04-14: 0.6174#29 2026-04-14: 0.6174#30 2026-04-14: 0.6174#31 2026-04-14: 0.6174#32 2026-04-14: 0.6174#33 2026-04-14: 0.6174#34 2026-04-14: 0.6174#35 2026-04-14: 0.6174#36 2026-04-14: 0.6174#37 2026-04-14: 0.6174#38 2026-04-20: 0.6174#39 2026-04-20: 0.6174#40 2026-04-20: 0.6161#41 2026-04-20: 0.6161#42 2026-04-20: 0.6161#43 2026-04-20: 0.6161#44 2026-04-20: 0.6161#45 2026-04-20: 0.6161#46 2026-04-20: 0.6161#47 2026-04-20: 0.6161#48 2026-07-10: 0.6161#49 2026-07-10: 0.6161#50 2026-07-10: 0.6010#51 2026-07-10: 0.6010#52 2026-07-10: 0.6010#53 2026-07-10: 0.6008#54 2026-07-10: 0.6008#55 2026-07-10: 0.6008#56 2026-07-10: 0.6008#57 2026-07-10: 0.6008#58 2026-07-10: 0.6008#59 2026-07-11: 0.6008#60 2026-07-11: 0.6008#61 2026-07-11: 0.6008#62 2026-07-12: 0.6008#63 2026-07-12: 0.6008#64 2026-07-12: 0.6008#65 2026-07-12: 0.6008#66 2026-07-13: 0.6008#67 2026-07-13: 0.6008#68 2026-07-13: 0.6008#69 2026-07-13: 0.6008#70 2026-07-14: 0.6008#71 2026-07-14: 0.6008#72 2026-07-14: 0.6008#73 2026-07-14: 0.6008#74 2026-07-15: 0.6008#75 2026-07-15: 0.6008#76 2026-07-15: 0.6008#77 2026-07-15: 0.6008#78 2026-07-16: 0.6008#79 2026-07-16: 0.6008#80 2026-07-16: 0.6008#81 2026-07-16: 0.6008#82 2026-07-16: 0.6008#83 2026-07-16: 0.6008#84 2026-07-17: 0.6008#85 2026-07-17: 0.6008#86 2026-07-17: 0.6008#87 2026-07-17: 0.6008#88 2026-07-17: 0.6008#89 2026-07-18: 0.6008#90 2026-07-18: 0.6008#91 2026-07-18: 0.6008#92 2026-07-18: 0.6008#1 2026-04-13#92 2026-07-18

Scientist findings — settled questions

ClaimConfidenceAgentDate
Adding venue-split (home_home_wpct/away_away_wpct), separated diff_off/def_rating, or symmetrizing form with away_win_pct_10 to the 16-feature star_dependency champion does NOT improve log loss (all statistical ties).highmeta_scientist2026-07-18
Promotion audit [vetoed] for TabPFN - trimmed 10-feature core (84a05669)highcritic2026-07-17
The 16-feature star_dependency champion can be trimmed to 12 features (dropping home_efg_pct, diff_net_rating, home_pie, away_pie) with no significant log-loss change — a valid production simplification, not an improvement.highmeta_scientist2026-07-17
Adding home/away roster_depth on top of the 16-feature star_dependency champion does not improve log loss (statistical tie); roster_depth adds nothing beyond star_dependency.highmeta_scientist2026-07-17
The 16-feature Ridge L2 (C=0.01) model with home/away star_dependency (0.615079) is the verified champion, replicated again and significantly better than the 14-feature anchor (0.616294, p=0.024); Optuna re-confirms C=0.01 is optimal for this exact feature set.highmeta_scientist2026-07-17
Promotion audit [vetoed] for TabPFN - trimmed 10-feature core (84a05669)highcritic2026-07-16

Recent runs

RunStartedStatusSummary
d9d4bb1d2026-09-01 14:51completedPipeline completed in 7s. Agents: 6, Total tool calls: 0
9f9559372026-08-31 17:36completedPipeline completed in 4s. Agents: 6, Total tool calls: 0
ffcbb0442026-08-30 14:52completedPipeline completed in 7s. Agents: 6, Total tool calls: 0
0020c3e92026-08-29 15:04completedPipeline completed in 7s. Agents: 6, Total tool calls: 0
45c891d22026-08-28 21:12completedPipeline completed in 6s. Agents: 6, Total tool calls: 0
a91146e42026-08-27 20:24completedPipeline completed in 5s. Agents: 6, Total tool calls: 0