manual_slop/conductor/directives/single_hypothesis_minimal_test/v1.md

Form a SINGLE hypothesis, test it MINIMALLY — do not bundle multiple fix attempts into one change

The rule

When debugging, the agent forms exactly one hypothesis at a time ("I think X is the root cause because Y"), states it explicitly, and tests it with the smallest possible change. The agent does not change multiple variables at once; the agent does not "fix several obvious problems" while in the debugging flow; the agent does not "while I'm here" improve unrelated code.

Per skills/systematic-debugging/SKILL.md:145-163 (Phase 3: Hypothesis and Testing):

Scientific method:

1. Form Single Hypothesis

   • State clearly: "I think X is the root cause because Y"
   • Write it down
   • Be specific, not vague

2. Test Minimally

   • Make the SMALLEST possible change to test hypothesis
   • One variable at a time
   • Don't fix multiple things at once

3. Verify Before Continuing

   • Did it work? Yes → Phase 4
   • Didn't work? Form NEW hypothesis
   • DON'T add more fixes on top

Why

Bundled fixes are unauditable. If the test passes, the agent doesn't know WHICH of the bundled changes was responsible; if the test fails, the agent doesn't know which to revert. The next fix attempt becomes a search across multiple variables, which is exponentially slower than testing one at a time.

Stating the hypothesis explicitly is a discipline against "I just tried a bunch of things." If the agent cannot state the hypothesis in one sentence, the hypothesis is too vague to test — refine first, then test.

What this means in practice

• Before each fix attempt, the agent writes: "I think [specific X] is the root cause because [specific Y]; I will test by [single change Z]."
• The change is single-line or single-parameter; not multi-file.
• If the change does not resolve the bug, the agent REVERTS the change (or notes it explicitly) and forms a new hypothesis. The revert is part of the protocol, not an option.
• "While I'm here" improvements are forbidden inside a debugging flow. The agent returns to the improvement in a separate commit after the bug is resolved.
• After three failed hypotheses (not three fix attempts in one commit; three SEPARATE hypotheses), the architecture question is reached (see three_fix_failures_question_architecture).
• The agent does not batch commits inside a debugging flow to "save time." Each hypothesis gets its own commit (atomic per hypothesis), with the test commit prepended when possible.