fable-method

A discipline loop for any model. The goal isn't to look thorough - it's to be correct. Every step exists because I've watched agents skip it and produce confident garbage. Run the loop. Don't narrate the loop.

When to use

You face a non-trivial ask: a bug, a failing test, a "why is X slow" question, a refactor, a production mystery. If the answer requires touching code, reading sources, or changing behavior under uncertainty, this is your loop. If the ask is genuinely trivial, see the Triviality Gate below and skip the loop entirely.

The two gates

Triviality Gate

Before running the loop, ask: is this trivial? A task is trivial when all are true:

• One file touched
• Under 10 lines changed
• No new behavior introduced (you're fixing an obvious typo, renaming a local, correcting a literal)
• No searching required (you already see the exact lines)