/ratchet:cut — the anti-delusion pass

Every plan rests on assumptions. Most are fine. A few are load-bearing and wrong, and those are the ones that waste days. This command finds them and hands you the cheapest test to kill each before it costs you.

Step 0 — Load state

ratchet status

Procedure

1. Surface the assumptions the current objective and chosen bottleneck depend on. Include the silent ones — "the API returns what the docs say", "the user actually wants this", "this file is the source of truth", "the test suite covers this path".
2. For each assumption, specify:
   • Breakage mode — what specifically goes wrong if it is false.
   • Detection signal — the observable that would reveal the break.
   • Falsification test — the cheapest action that would prove it false. Prefer a five-minute check over a five-hour build.
3. Rank by damage × likelihood. An assumption that is cheap to test and catastrophic if wrong goes first.
4. Pick the top three kill-tests — the ones worth running before any further build. A kill-test is a test designed to end the plan cheaply, not to confirm it.

Output contract

ASSUMPTIONS (ranked by damage × likelihood):
- <assumption> | breaks: <mode> | signal: <detection> | test: <cheapest falsification>
- ...

TOP 3 KILL-TESTS (run before building):
1. <test> — kills the plan if: <result>
2. ...
3. ...

Serialize

Record the riskiest untested assumptions so the loop can't forget them:

ratchet state append assumptions '{"text":"...","killTest":"...","status":"untested"}'
ratchet state set phase cut

Untested assumptions drain confidence (ratchet score confidence). Run the kill-tests, then mark survivors: append with "status":"tested" or "status":"killed".

Next: if the plan survives, /ratchet:build. If a kill-test lands, /ratchet:lock again.