wrote tune-the-cache.eval.yaml, tune-the-cache.task.md (1 case)
mapped:   task -> prompt_file
mapped:   2 criteria -> rubric (weights kept as text)
assumed:  no pass threshold; the eval default 0.70 applies
assumed:  expect_trigger: true (the scenario targets a skill)
unmapped (6), AR9A5:
  $.agent                      value "2 item(s)"  -> use `eval run --harness` and `--model`
  $.baseline                   value "2 item(s)"  -> use `eval run --ablation`
  $.criteria[0].category       value "behaviour"
  $.labels                     value "2 item(s)"
  $.owner                      value "platform-team"
  $.repeats                    value "3"  -> use `eval run --runs N`
