Bounded decisions

1. Decide whether it is worth a decision. Use this before substantial research, after repeated failures, before loading several tools or skills, before spawning agents, when routes differ materially, or before a consequential proposal. Skip simple answers, calculations, routine edits, and choices that cannot change the next step.
2. List 2–4 bounded options. Each is one concrete next step you could start now, not a goal. Write each in one line.
3. Cost and reversibility. For each option, state what it costs (time, tokens, money, other people's attention) and how hard it is to undo: free to undo, undoable with work, or irreversible.
4. Evidence. For each option, state the observed facts that favor or count against it: a command's output, a file and line, a failing check, a prior attempt. Mark anything you have not checked as an assumption.
5. Pick one. Prefer the option the evidence supports best at the lowest cost that can still be undone. Between two options that are close, take the cheaper, more reversible one first.
6. Record why in one paragraph: the options, the one you chose, the evidence that decided it, and what would make you switch. Put it where the work is recorded: the task file, the decision record, the pull request description, or your report.

Escalate to the human only when every viable option is irreversible, or when the choice falls outside the task's scope. Otherwise proceed on your pick; the human can course-correct afterward.