Merit engine & OzmaScore
Transparent rankings from reliability, adoption, retention, value, and momentum.
View as Markdown →Ozma ranks APIs with objective merit signals derived from real proxy usage — not ads or manual curation. Scores roll up every hour from usage_events into provider_stats and score_snapshots via the merit engine in packages/merit.
Five axes (30-day window)
| Axis | Signal |
|---|---|
| Reliability | Success rate × latency factor (Wilson lower bound) |
| Retention | Repeat payers / unique payers (Wilson; min sample thresholds) |
| Adoption | GMV + payer count (log-scaled; shrinkage on low volume) |
| Value | Successful calls per dollar, adjusted for refunds |
| Momentum | Paid calls in last 24h (hotness curve) |
Small samples shrink toward neutral priors so new APIs cannot instantly outrank established ones with sparse data.
OzmaScore (0–100)
Category-normalized percentiles of each axis blend into a weighted score (reliability 30%, retention 25%, adoption 20%, value 15%, momentum 10%). Each API gets rank_in_category on snapshots.
Search & leaderboards
Six public leaderboards — top_grossing, most_used, most_reliable, best_value, trending, new_rising — refresh on cron. New & Rising uses Thompson sampling for low-data APIs with reserved exploration slots.
When you search with a query, default sort relevance blends semantic match (~75%) with OzmaScore (~25%) among relevant candidates (merit ranking v2, default ON).
Health & anti-gaming
Daily health probes suspend APIs after three consecutive upstream failures — suspended listings drop from merit rollups. Self-dealing (same org as consumer and provider) is excluded. Payer concentration caps reduce merit when one consumer dominates an API's usage.
Rankings are earned from marketplace behavior — transparent on every listing and API detail page.