Async adversarial multi-model code review system. Parallel LLM fan-out, debrief synthesis with convergence analysis, severity-gated merge blocking. 106 tests, Python.

Same spec as Pidgeon, rebuilt in 1 hour using design-first parallel agent execution. 55 tests, 17 commits. A direct comparison of process vs velocity.

Carrier-agnostic shipping rate integration. Outside-in TDD, 48 cross-model adversarial reviews, machine-queryable decision log. 161 tests, 86 commits.

Multi-agent AI evaluation platform. 1,300+ commits across 3 phases, 1,503 tests, 508 test files, ~28k LOC. Structured contests between agent configurations with observable traces, scoring, failure tagging, and cost visibility.