Concepts

These pages explain why TokenFuzz is shaped the way it is. They describe the system rather than walking through an operator task.

If you are reading them in order, start with Audit lifecycle for the end-to-end story. The other pages explain individual components and cross-cutting design choices.

Page	What it covers
Audit lifecycle	A run from setup to a reviewed finding or maintainer crash bundle, in one place, with a diagram.
System architecture	The components: audit run, work queue, agents, probe runner, triage, backends.
Strategy model	The eight investigation methods, how cards get a strategy, and how evidence-aware rotation works.
Cost model	What scales with cost on long runs, and the levers the harness gives you.
Deduplication	How crashes (stack-similarity clustering) and findings (evidence clustering) group related reports for review.
Benchmarking	How bin/benchmark compares TokenFuzz against a direct prompt without hiding orchestration or review cost.