A loop should state what changes each iteration, how progress is measured, and when budget, success, or futility stops work. Repetition alone does not create improvement.

Loops Need Progress and Stop Conditions

Beam search keeps only a fixed number of promising partial solutions at each depth. It controls cost but can permanently discard the route to the best solution.

Beam Search Trades Coverage for Bounded Width

Time, tokens, money, tool calls, and risk tolerance are finite resources. Carrying remaining budgets in state allows routers and loops to degrade or stop predictably.

Budgets Are First-Class Graph State

Deadlock occurs when participants wait cyclically; livelock occurs when they keep reacting without progress. Timeouts, lock ordering, progress measures, and arbitration break these states.

Deadlock and Livelock Need Recovery

A reflection step critiques or revises an output, but self-review can repeat the same blind spot. Ground it in criteria, tests, retrieved evidence, or a differently informed reviewer.

Learn Before

Related

Related