probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (drop)
  probing retrieve_the_medicine (baseline)
  probing retrieve_the_medicine (drop)
  probing mutational-doctor-no-such-scenario (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)
  probing tidy_the_kitchen (baseline)

==================================================================
DOCTOR  is this adapter ready to measure anything?
==================================================================

  ok    the adapter answers          4 runs, median 31ms
  ok    a baseline passes            tidy_the_kitchen
  ok    harness failure is visible   an unknown scenario reported a harness error
  ok    the canary is caught         tidy_the_kitchen
  note  a fault can be refused       nothing came back inapplicable in this probe
        Not necessarily wrong - the probe is small. But an adapter
        with no refusal path scores faults it silently approximated,
        and those enter the denominator as survivals.

  ok    repeated runs agree          tidy_the_kitchen twice at one seed: pass
  ok    concurrent runs agree        tidy_the_kitchen passed 8 times at once
  ok    a detection is possible      --trials 5 is enough to tell a fault from 5 clean baselines

  a full campaign  2 scenarios, 82 defects between them (homebody orchestration@1)
                   at most 10 baseline + 10 canary + 410 defect trials
                   = 430 runs at ~34ms each: about 14.7s serially, 1.8s at --workers 8
                   That bounds the runs that are evidence, and early stopping
                   lowers it. A run that errors or is refused is not a trial and
                   comes on top: 3 for a defect that is never delivered, and
                   for one that lands in some placements only, its trials again
                   at most.