ablatelisted

# /ablate - one-sided harness ablation ## Input `$ARGUMENTS` is an element path narrowing the measurement to one element. Omitted, every element Phase 1 enumerates is measured. ## Where the criteria and thresholds live The arm list, the run count per arm, and the pass threshold are all constants in `${CLAUDE_SKILL_DIR}/scripts/arms.py`. The classification criteria live in `${CLAUDE_SKILL_DIR}/scripts/verdict.py`. The DR-gate criteria, the confirmed-unmet marker and where the records are read from, live in `${CLAUDE_SKILL_DIR}/scripts/dr_gate.py`. The measurement window and the rare-by-design set live in `${CLAUDE_SKILL_DIR}/scripts/usage_counts.py`. The per-rule trigger tasks and the fixed task set live in `${CLAUDE_SKILL_DIR}/references/measurement-criteria.md`. Do not copy a number into this body (`docs/wiki/deterministic-script-judgment.md`). ## Phase 1: Enumerate Call `enumerate_elements(root)` in `skills/_lib/harness_elements.py` for the harness elements and each one's classification. When `$ARGUMENTS` names an element path, hand Phase 2 that one alone. ```bash python3 -c 'import sys; sys.path.insert(0, "skills/_lib"); import harness_elements, json; print(json.dumps(harness_elements.enumerate_elements(".")))' ``` ## Phase 2: Run the arms For each element Phase 1 returned, and each arm in `arms.ARMS`, assemble the command `arms.arm_command(arm, element)` returns and run it `arms.RUN_COUNT` times. Take that element's triggering task from `${CLAUDE_SKILL_DIR}/refer