Lesson 16 / 25

A Catalogue of Loop Failures

Name the common failures and match each to the guard that catches it.

Name it to fix it

Repetition: the same call again and again. Thrashing: trying many different fixes that all fail. Drift: wandering away from the original goal. Context bloat: history grows until quality drops. Silent overspend: every step is fine but the total is huge. Each has a different guard, so logging which one happened tells you what to improve.

What goes wrong

Most loop incidents are one of a few patterns: repeating, thrashing on errors, or quietly overspending.

Four common patterns, each with its own guard.
Figure 5.1 — Repeat, thrash, drift and overspend.

Failure to guard map

Use this table as a checklist when reviewing a loop.

Failure          Guard
Repetition       RepeatGuard on (tool, args)
Thrashing        cap consecutive tool errors; escalate
Drift            restate goal each N steps; verify against goal
Context bloat    clip results, summarise, offload
Silent overspend total token and cost budget

Quick check: Every step is cheap but the run costs a lot overall. Which failure is this?

  • Repetition
  • Silent overspend
  • Drift
  • Syntax error
Answer

Silent overspend — Only a total budget notices cost that accumulates across many individually fine steps.