Forward Track Record

Every prediction is logged before its horizon and graded after it matured — no backtests, no survivor editing, control group included. Machine-readable: GET /api/track. Generated 2026-09-02T12:48:59.

Honest verdict (verbatim)

Long-horizon graded: 73 calls, hit-rate 0.808, discrimination -0.037 (high- minus low-confidence bucket; >0 = confidence carries real information). Directional layer: the maturing 14d cohort is event-occurrence calls WITHOUT inherent direction - it is excluded from directional stats by design; raw signed ret_pct is recorded for transparency.

Canonical early warning (14-day horizon)

1bets open
12bets graded
11control open
92%bet hit-rate

First maturity: 2026-09-07T13:47:28. Bets = high-conviction canonical calls; the control group are deliberate low-conviction calls ("these stay quiet") — discrimination between the two is what proves skill, not hit volume.

Short-horizon calls (<7 days, autopilot)

6832graded
32.9%hit-rate
0.508Brier
149open
confidence bucketnhit-rate
niedrig1040%
mittel143245%
hoch539030%

Hit = notable market move on the entity within the horizon (movement proxy, not directional P&L). Brier: lower is better; 0.25 = coin flip at 50% confidence.

Directional layer (parallel, additive)

32directed calls
6874excluded (undirected)
6.89%mean ret
3036with price path

Calls without an inherent direction are EXCLUDED, not fabricated. Raw signed returns are reported for transparency; any claim that they beat chance must first be registered in the multiple-testing register.

Council verdicts (48h forward grading)

950total
84graded
62%hit-rate
0.697trust factor

Method