{
 "format": "LSC-1",
 "lsc": "LSC-1",
 "edition": "E1",
 "cut": "2026-10-09",
 "name": "THE EVAL CARD",
 "handle": "eval-card",
 "number": "UNSEATED",
 "house": "THE INSTRUMENTS",
 "wing": "THE APPLIED-AI WING",
 "role": "evals that settle the argument",
 "tagline": "A pass rate with its interval, and a change that must beat chance.",
 "sigil": "◆",
 "hue": "#4A3BC4",
 "punch": "86FF60FBFF0B2D63A94500A2",
 "punch_of": "SKILL.md · sha-256 · first 24",
 "score": 98,
 "tier": "LEGENDARY",
 "layer": "WITNESSED",
 "score_line": "SELFTEST 14/14 · FOUR COMMANDS ON THE RECORD",
 "honest_line": "I'm an AI reading a card. I can be wrong — I can invent a number, a rule, a name or a date — so check anything that matters against your own records, the source and a professional before you act.",
 "term": "one purchase · yours to keep · nothing renews itself",
 "wakes": "someone asks to test, grade or compare a prompt, model or skill, or to prove a change helped",
 "loop": [
  "write the eval set from real work",
  "run the outputs, then evalkit.py grade",
  "judge the rubric cases, never guess them",
  "compare: fixed, broken, and is it real?",
  "Write the ledger: score · interval · why"
 ],
 "laws": [
  "**The eval set comes from real work**: real inputs, the right answer written down, the hard and rare cases on purpose — never invented from what the model already says.",
  "**One check per claim.** A check the machine can run beats a judgement; a judgement beats a guess.",
  "**A rubric case is judged, never guessed.** Until a judgment is written it is counted apart, never as a pass.",
  "**Every pass rate travels with its interval.** Under 30 graded cases, say the set is small.",
  "**A change wins only by more than chance**: the cases it fixed must outnumber the ones it broke, by the sign test at p < 0.05.",
  "**When Claude judges**, it judges against the rubric alone, strictly, one line of reason each. Where the stakes are high, a person judges.",
  "**Keep every run's results.** The history is the regression suite."
 ],
 "never": [
  "count an ungraded case as a pass",
  "call two flipped cases a win",
  "build the eval set from the model's own output"
 ],
 "machine": "scripts/evalkit.py · new · grade · judge · compare · selftest",
 "proof": "selftest 14/14 · FOUR commands on the record",
 "faces": [
  "card.svg",
  "card-machine.svg"
 ],
 "box": [
  "SKILL.md",
  "CARD.md",
  "TIN.md",
  "START-HERE.md",
  "LICENSE-NOTE.md",
  "THE-CLAUSES.md",
  "MY-LEDGER.md",
  "WITNESS.md",
  "SELFTEST.md",
  "AUDIT.md",
  "MANIFEST.md",
  "HOSTS.md",
  "scripts/evalkit.py",
  "references/WRITING-CASES.md",
  "references/READING-A-SCORE.md",
  "rules/ (4 doors)",
  "system/ (2 prompts)",
  "mcp/mcp_card.py",
  "card.json",
  "card.svg",
  "card-machine.svg",
  "card.html"
 ],
 "hundred": "Trigger 5 · Machinery 5 · Law 5 · Portability 5 · Proof 5 — 25 of 25 → 98 of 100 · LEGENDARY"
}