Explanation

This page lists the explanations, which cover why the library is built the way it is.

• How the library is built walks through one monitored step, the protocols and their outcomes, what the monitor reads, how each monitor scores, task authorship, halts, subagents, records, streams and traces, graph steps and errors, and ends with every known limit.
• Where the ideas come from credits every paper and codebase the library draws on.
• Live runs of a monitored agent reports a small evaluation with real models: an agent under each monitor and three of the protocols, once honestly and once with a hidden side task, step by step.