Forecast ledger scoreboard

Calibrated Health Agent · updated 2026-11-06T00:32 UTC · days locked 14 · scored 10 · void 0 · no journal 4

Every morning seven forecasters bet on one person's evening mood, energy and soreness (1–10). The bets are locked and hash-chained before the journal exists, then scored with the ranked probability score (lower is better). Skill above 0 beats the reference; below 0 loses to it. Results are published as-is, including forecasters that lose to persistence, and the real n is shown next to every number. The chain hashes are in chain.csv.

Primary endpoint

gemini:gemini-3.5-flash-lite vs persistence, energy skill: +0.15 (95% interval -0.19 to +0.40, n = 10)

Secondary endpoints

ForecasterReferenceTargetPaired n RPS skill (95% interval)
gemini:gemini-3.5-flash-litepersistencemoodn = 10-0.12 (-0.12 to +0.24)
gemini:gemini-3.5-flash-litepersistencesorenessn = 10-0.05 (-1.30 to -0.02)
gemini_notes:gemini-3.5-flash-litegemini:gemini-3.5-flash-liteenergyn = 9n below 10

All forecasters

ForecasterTargetScoredMean RPS Skill vs climatologySkill vs persistenceMean log score
climatologyenergyn = 100.100–0.29 (n = 10)2.050
climatologymoodn = 100.090–-0.04 (n = 10)1.940
climatologysorenessn = 100.080–-0.37 (n = 10)1.680
gemini:gemini-3.5-flash-liteenergyn = 100.120-0.20 (n = 10)0.15 (n = 10)2.150
gemini:gemini-3.5-flash-litemoodn = 100.100-0.08 (n = 10)-0.12 (n = 10)1.900
gemini:gemini-3.5-flash-litesorenessn = 100.0600.24 (n = 10)-0.05 (n = 10)1.570
gemini_notes:gemini-3.5-flash-liteenergyn = 90.0900.13 (n = 9)0.42 (n = 9)1.790
gemini_notes:gemini-3.5-flash-litemoodn = 90.110-0.10 (n = 9)-0.08 (n = 9)1.980
gemini_notes:gemini-3.5-flash-litesorenessn = 90.0600.25 (n = 9)0.07 (n = 9)1.550
persistenceenergyn = 100.140-0.41 (n = 10)–3.830
persistencemoodn = 100.0900.04 (n = 10)–2.690
persistencesorenessn = 100.0600.27 (n = 10)–2.530
seasonal_naiveenergyn = 50.170-0.35 (n = 5)0.18 (n = 5)6.000
seasonal_naivemoodn = 50.160-0.62 (n = 5)-0.12 (n = 5)4.930
seasonal_naivesorenessn = 50.0600.17 (n = 5)-0.02 (n = 5)1.470

Pre-registered analysis plan: docs/ledger/analysis-plan.md in the project repository; its hash is the first line of chain.csv. Wellness forecasting, not medical advice.