A decision-forcing loop for UX, tooling, library, or architecture calls where the failure mode is handing back an option menu — or dismissing a candidate from training priors. Each pass states the explicit criteria, then RESEARCHES one candidate with real evidence (docs, source, a quick spike or benchmark) and scores it, eliminating the weakest each round until exactly one remains. It exits only with a single committed pick plus a one-paragraph rationale naming the runner-up and why it lost — never a tie, never a 'here are three options' hand-off.
Loop
- Goal
- a single committed recommendation among the candidate options, each researched against explicit criteria with real evidence, ending in a written rationale and a named runner-up
- Exit when
- decision.md exists with zero unchecked '- [ ]' candidates, exactly one pick scored against the stated criteria, and a rationale naming the runner-up
Check between iterations
test -f decision.md && test $(grep -c '^- \[ \]' decision.md) -eq 0 && echo READY || echo NOT_READY
$
Kickoff prompt
Start the "Decisive Pick Loop" loop.
Goal: a single committed recommendation among the candidate options, each researched against explicit criteria with real evidence, ending in a written rationale and a named runner-up
Max iterations: 5
Between iterations run: test -f decision.md && test $(grep -c '^- \[ \]' decision.md) -eq 0 && echo READY || echo NOT_READY
Exit when: decision.md exists with zero unchecked '- [ ]' candidates, exactly one pick scored against the stated criteria, and a rationale naming the runner-up
Step 1: Write decision.md with the explicit decision criteria and a '- [ ]' checklist of candidate options, then research the first candidate with real evidence (not priors) and score it against the criteria.
›