Definitions

Point fields (from /api/metrics)

• t: ISO timestamp (bucket start, hourly for 24h/7d, daily for 30d)
• costUSD: Sum of Run.costUSD in bucket (rounded to 5 decimals)
• errorRate: errors / count with status != 'success'
• p95LatencyMs: 95th percentile of latencyMs in bucket
• runs: total runs in bucket
• tokensIn: sum of inputTokens
• tokensOut: sum of outputTokens

Aggregates

• costByProvider: sum of costUSD by provider
• costByModel: sum of costUSD by model

Reliability metrics

• Error rate: fraction of non‑success over total
• Success rate: 1 - errorRate

Percentiles

• p50/p95/p99: compute over the set of latencyMs values in the bucket.
• Current API returns p95LatencyMs only.

Cost normalization

• Cost per 1k tokens (example): costUSD / ((tokensIn + tokensOut) / 1000) when tokens > 0.

Examples

{ "t": "2025-01-01T01:00:00.000Z", "costUSD": 0.12345, "errorRate": 0.01, "p95LatencyMs": 850, "runs": 42, "tokensIn": 1200, "tokensOut": 800 }