What the times measure
| Field | Details |
|---|---|
Response time | From calling the API to receiving its response. Includes upload, network and service work. |
generation_ms | Server generation only. Does not include the full HTTP exchange. |
P95 | 95% of measured requests completed at or below this time; 5% took longer. |
Generation success | API HTTP 200 with complete mode-specific telemetry: headers, or form_fields and form_body. |
Target acceptance | The target accepted the subsequent application request. Check this separately. |
Measured scope
The homepage’s ~7 ms figure comes from a 16 October 2026 warm-origin HTTPS benchmark: 60/60 successful generations, 7.24 ms mean, 6.71 ms median and 9.52 ms P95. The client ran on the origin host; this excludes internet transit from your location. It is a historical warm-path measurement, not an overall production average or a capacity test.
The 99.9% generation-success target is not a measured overall production success rate, an uptime SLA or a target-site acceptance rate. First-time processing, changed material and unsupported contexts may take longer.
A 13 October 2026 test completed 2,700/2,700 generations at 15 aggregate requests per second, averaging 386.729 ms. It used four selected warm material references and the twelve-slot deployment. It is a historical benchmark, not a measurement of current tier capacity or an SLA for every script or client.
- First uploads may take longer than reference reuse.
- Target acceptance also depends on current material, profile, cookies, body and transport.
- The account rate ceiling is not reserved capacity. Shared service capacity uses short bounded burst waits; sustained overload can return 429 or 503. Honor
Retry-After.