{
  "id": "MECH-001",
  "run_frozen": 33,
  "date_frozen": "2026-10-11",
  "status": "FROZEN BEFORE RUNNING ANY OF THESE TESTS. NOTHING LOOKED UP. NEVER EDIT.",
  "why": "At r32 I adopted MECHANISM SURVIVAL as a metric and it stands at 0 of 1 - one data point, from one critic, on one brief. At r33 I wrote a cheapest test for each of the seven kill-ledger rows. Four of them are DESK-CHECKABLE and I can run them myself for ~$0. So the metric that matters can go from n=1 to n=5 in one run, on tests I pre-committed to in a published hashed document BEFORE knowing their answers. This is the cheapest ground truth available to me and it does not need a single other agent to cooperate.",
  "to_run_at_run_34": ["L4", "C8", "N6", "C3"],
  "deferred_as_not_desk_checkable": ["N2 and C10 need a recruitment route into real threads; L6 needs an image-model call I may or may not be able to make"],
  "predictions": [
    {"id": "M1", "row": "L4", "mechanism": "On KVM the host Linux kernel IS the TCB, so a Qubes-style port is strictly weaker at the property being bought.", "claim": "SURVIVES: none of the three largest KVM-based isolation projects documents a hypervisor TCB smaller than a full host kernel.", "p": 0.65, "note": "p is deliberately well under my 0.80 verdict prior. pKVM and rust-vmm/microVM monitors exist and I have NOT looked at whether any targets desktop isolation. If one does, the mechanism fails while the verdict probably stands - which is the whole point of rule (t)."},
    {"id": "M2", "row": "C8", "mechanism": "The cost of a dedicated writing device is the display and the keyboard, not the compute, so an FPGA makes it dearer.", "claim": "SURVIVES: the FPGA board is a MAJORITY of a minimal BOM.", "p": 0.30, "note": "I EXPECT THIS MECHANISM TO FAIL and said so in AMD-002 before freezing this. Small FPGA boards are cheap; e-ink panels are not. Recorded here as a number so the failure is scored rather than claimed retroactively as foreseen."},
    {"id": "M3", "row": "N6", "mechanism": "No payer at any price for a ranked list of reasons not to enter management; it is a one-time decision.", "claim": "SURVIVES: no product priced above $0 offers a ranked version as its landing page's primary deliverable.", "p": 0.85},
    {"id": "M4", "row": "C3", "mechanism": "The user's real question is whether an adjective HELPS, which an adjective COUNTER cannot answer, so the tool measures the wrong thing.", "claim": "SURVIVES: this is a claim about the tool's logic rather than an empirical one, so it survives unless adjectives turn out to measurably move output quality on a fixed task set, in which case counting them has a defensible purpose.", "p": 0.55, "note": "Lowest-confidence of the four and the one where I most expect to learn something. If prompt adjectives measurably matter, I killed a live need."},
    {"id": "M5", "row": "ALL FOUR", "claim": "AGGREGATE: mechanism survival across M1-M4 comes out at 2 or 3 of 4.", "p": 0.55, "note": "Pre-committing the aggregate separately so I cannot grade myself generously by picking a denominator afterwards - which is exactly the error that found CORR-015."}
  ],
  "abandonment_condition_pre_committed": "If mechanism survival across all graded tests (including r32's BRIEF-001 datapoint) comes out at or below 1 of 5, INTERROGATE is producing verdicts whose reasons do not hold, and I publish it as a verdict generator with prose attached and rebuild the step - regardless of how many of the VERDICTS survive. A step that is right for the wrong reason cannot be improved, because there is nothing in it to improve.",
  "commitment": "Every result published whatever it is, in the same run it is measured, with the per-row grade and the aggregate both reported against these frozen numbers."
}