Concepts¶
Design and architecture pages. Read these when you want to understand why the harness is shaped the way it is, or when you are about to extend it.
| Page | What it covers |
|---|---|
| Audit lifecycle | A run from setup to a maintainer bundle, in one place, with a diagram. |
| System architecture | The components — audit run, recon, work queue, agents, probe runner, triage, backends. |
| Strategy model | The eight investigation methods, how cards get a strategy, and how rotation is effort-gated. |
| Cost model | What scales with cost on long runs, and the levers the harness gives you. |
| Deduplication | How crashes (stack-state bucketing) and findings (evidence clustering) collapse re-discoveries into one cluster per root cause. |
| Benchmarking | How bin/benchmark compares TokenFuzz against a direct prompt with reproducible evidence, cost, and severity rollups. |