ActionBench
Reliability of orsel actions per site, action and recipe version, with failures published.
ActionBench is the public record of how reliably orsel actions run. Every number comes from our own runs: scheduled canaries on synthetic inputs, and client runs from people who opted in to share outcomes. Read and write actions are reported separately, and failures are published with the rest. How the numbers are counted is on the methodology page. The raw data is at /v1/benchmark as JSON, and it is the source of this page.
Generated 2026-11-18T02:06:29.797Z.
reliability
No runs have been recorded yet, so there are no scores to show.
prompt injection suites
No suite has been run and published yet.