Status and performance. Measured, not promised.
Live health, latency and success rate for every endpoint, measured on production from customer requests and our scheduled checks.
Live health
Checking live health…
Health comes from scheduled checks with known public inputs. Where there is enough traffic, the share of real requests that succeeded in the last hour is shown next to it.
- Measured on
- —
- Window
- —
- Requests measured
- —
- From scheduled checks
- —
Most of these requests are our own scheduled checks. Customer traffic is still small, so the numbers mostly show how each endpoint answers known public inputs.
The numbers on this page load live from /status.json, our status document, so they are never older than it. Turn on JavaScript to see them here.
Latency and success rate, per endpoint.
p50 and p95 latency of successful responses, the share of requests that succeeded, and how many requests each number rests on. Failed requests are never charged, whatever the success rate.
“—” means the endpoint had too few requests in the window to publish a number.
How these numbers are measured.
- Which requests. Requests to production in the window shown above: customer requests plus our own scheduled checks, which call every endpoint with known public inputs. The page shows how many came from the checks. Load tests are left out, and so are cache hits and requests turned away before any work was done (
unauthorized,forbidden,insufficient_credits,spend_limit_reached,rate_limited,request_too_large,method_not_allowed,endpoint_retired,idempotency_key_reused,idempotency_key_in_progress,job_not_foundandunknown_endpoint). - What counts as success. Any correct answer. A request fails when the platform or SocialTap could not produce one (
upstream_blocked,upstream_errorandinternal), or when a scheduled check got a wrong answer. Errors that correctly describe the request or the item (invalid_input,invalid_cursor,not_found,private_or_unavailable,no_transcript,no_live,no_mediaandoverloaded) count as successes: the API did its job. - Latency. p50 and p95 are the time SocialTap took to produce successful answers: half of them were faster than p50, 95% faster than p95.
- Always current. The numbers are recomputed every few minutes and load live from /status.json each time this page opens.
- Sample size. Each row shows how many requests it is based on; endpoints with too few requests show no numbers.
- Rounding. Success rates are rounded down, so 99.96% shows as 99.9%, never 100%.