This page is the single source of truth for the rules that govern every number displayed on ALZ. If any value on the ranking doesn't match these rules, it's a bug — report it.
Client timestamps are informative only. The moment the ingest endpoint receives the request is the sole time of record.
Predictions received after the bucket-minus-10-seconds mark are rejected with 422. No exceptions, no retroactive accept.
Each (algo, bucket) pair is unique. Predictions cannot be updated, replaced, or deleted — only resolved.
Every algorithm displays exactly two measurements — bets per day (volume) and accuracy with its Wilson 95% interval (quality). Nothing else. No confidence scores that fold both into an opaque number, no ELO, no Sharpe — anything derived is auditable on the ranking table without taking our word for it.
The Wilson interval is the canonical way to express binomial uncertainty. A tighter interval means more bets, a wider one means fewer. When the interval is wide, the algorithm is visually de-emphasized on screens so the viewer learns by osmosis that 67% ± 4 is serious and 67% ± 18 is not.
One number per algorithm. Combines edge (accuracy) with volume so an algo with a tight-but-rare signal can't outrank a steady, high-volume one. The score is computed on the rolling 7-day window.
The discontinuity at 55 % is deliberate. It refuses to interpolate between "loses money long-term" and "makes money long-term" — they're different categories, not different shades. The volume tiers exist so that an algo on 2 bets/week can't sit next to one on 200 bets/week just because both happened to be correct.
Every algo enters the public ranking immediately. Score honestly reflects its sample — under 1 bet/day caps the score at 58, sub-55 % accuracy caps it at 55. No quarantine.
Dropped below breakeven for 30 consecutive days. Ledger preserved, no new predictions accepted.
The engine is live: predictions are written to a public Postgres ledger at every 5-minute Binance close, then resolved against the canonical close price. Algorithms launched today appear in the ranking immediately with a score that honestly reflects their thin sample — no quarantine, the scoring v5 caps young algos until they accumulate volume. Accuracy figures stabilize as the ledger grows.