heldout-snapshot.json

{
  "_README": [
    "NON-REGRESSION BASELINE for heldout.json — machine-checked by heldout.test.ts.",
    "",
    "An id set to `true` has passed at least once, so it may NEVER start failing again: that is the",
    "single hard rule of the held-out set (there is deliberately no accuracy threshold). Flipping one",
    "back to false, or deleting it, is only legitimate as a reviewed diff in this file with a reason",
    "in the commit message — which is the whole point of committing the baseline instead of computing",
    "it at run time.",
    "",
    "A NEW query starts absent from this map. It is reported but not enforced until it passes once;",
    "add it here in the same commit that makes it pass."
  ],
  "recorded": "2026-07-25",
  "passing": {
    "ho-001": true,
    "ho-002": true,
    "ho-003": true,
    "ho-004": true,
    "ho-005": true
  }
}