CW-301c · Module 1

Stage Gating & Error Recovery

4 min read

A stage gate is a checklist that the output of one stage must pass before the next stage begins. Without gates, errors propagate downstream — a research stage that misses a key competitor produces a draft that omits that competitor, which produces a review that does not catch the omission because the reviewer does not know what was missed. The error compounds at every stage until the final deliverable is wrong in ways that are invisible from within the pipeline.

The gate prompt pattern is explicit: "Review this [stage output] against the following criteria: [list]. For each criterion, respond PASS or FAIL with a one-sentence explanation. If any criterion fails, list the specific corrections needed before proceeding." Claude is thorough when given explicit criteria. It is unreliable when asked to "check if this looks good." Specificity is the gate. Vagueness is the hole in the gate.

Do This

  • Define 3-5 explicit pass/fail criteria for every stage gate
  • On gate failure, loop back to the current stage with specific corrections
  • Log gate failures — patterns reveal systemic prompt weaknesses that need fixing

Avoid This

  • Ask Claude to "review this" without specific criteria — it will say it looks fine
  • Skip the gate when you are in a hurry — that is when errors are most likely
  • Forward gate failures to the next stage hoping it will self-correct — errors compound