Prometheus exposition format metrics. Public, no auth, no credits. text/plain content-type.

Industry-standard scrape target: customers' Prometheus, Grafana Agent, Datadog OpenMetrics integration, or any OpenMetrics-compatible scraper polls this endpoint every 15-60 seconds. The metrics include:

• parlayapi_requests_total{status_class="2xx|3xx|4xx|5xx"}
• parlayapi_endpoint_requests_total{path, status_class}
• parlayapi_endpoint_request_duration_seconds{path, status_class}
• parlayapi_worker_uptime_seconds
• parlayapi_api_version (gauge with version_string label)
• parlayapi_postgres_deadlocks_total (always 0 since iter_8)
• parlayapi_idempotency_keys_active (currently 0 if Redis miss)

All metrics carry a worker label so customers' Prometheus deduplicates correctly across the 5-worker pool. Customers aggregate across workers via sum by(status_class) queries.

Cache-Control: no-cache (always fresh). The body is small (~2 KB) so the cost is trivial.

sleep_iter_40 #539.