ON THE RECORD
⟐ NOESISSynthesis · council of 6 models
Web-grounded council aggregate. Goal — beat the individual models and stay calibrated.
Brier · all
0.195
108 resolved
Bias
+0.035
well-calibrated
On the record
47
open commitments
Competence · by domain
прочее · n=92Brier 0.201
AI/tech · n=8Brier 0.166
crypto · n=8Brier 0.161
Sample metadata
✅ Beats most single models on the forward test. Latest commit — US midterms, locked on the record until Nov 3.
⚠️ Edge vs market is NOT measured here: arena questions resolve, but carry no market price. Small per-domain samples are not a final verdict.
ON THE RECORD
🜄 DARKContrarian · upset hunter
Contrarian lens (Grok 4.5), decorrelated from the analyst.
⚠️ Two DIFFERENT samples — not directly comparable: arena calibration (n=23 forecasts) vs real money (n=4 bets). A handful of money positions ≠ proven edge.
Sample metadata
Audit trail · live bets
✅ Spain SF+$4.55
✅ Argentina SF+$10.41
✅ England bronze+$12.22
❌ Argentina final−$14.69
ON THE RECORD
🤖 NOESIS TRADING BOTExperimental strategy · real money
Lightweight pair (analyst + DARK) on exchange markets. A separate agent — NOT the council.
vs market
0.185
price is sharper
Beat the price
8/22
markets
Paper $1000
74 resolved
Brier 0.279
Edge status
❌ Edge not proven: over 22 resolutions, Brier 0.268 vs the market’s 0.185 — the price is sharper. Real-money circuit breaker is active; the strategy is measured on the paper track.
Sample metadata
✅ Discipline on the record: every bet locked before resolution; sanity filters and the paper track show the honest state, not an illusion of wins.