Search the runs

What can be asked

Each term is a field the record names; every term must match. Values are matched exactly, as recorded.

  • setting:<step>.<name>=<value>: a step's recorded setting; * for any step
  • input:<name>=<digest>: a step's input by name and digest
  • inputs:same-as:<run>: every input of the first step equal to that run's
  • first_diff:<step>: a held comparison of the run first differs at that step
  • first_diff:any:<kind>: ...and a setting, input, configuration or output differs there
  • diff:same-as:<run>: first differs from the same baseline, at the same step, in the same way
  • verdict:<word>: the verifier's verdict, PASS or FAIL
  • row:"<name>"=<result>: a row of the verifier's that read FAIL or NOT-RUN
  • model:<field>=<value>: a model-call record's field: model, weights, engine, engine_version, dtype, batch_invariant
  • engine:<name>: a model-call record's engine
  • gated:<step>: a step the manifest declares gated
  • pipeline:<id>: the console's pipeline, P1, P2, ...
  • since:<run>: runs that arrived after that run
  • benchmark:yes|no: marked as a benchmark, or not

Compared with pz-59-agent-capture-baseline, the run this search was asked from.

No run matches every term.