← Back

Blog

13 Phases. 63 Decisions. 138 Bugs in the Spec.

11 min

An adversarial code review protocol — designed by one model, scrutinized by every model that didn't build it.

2026-04-04

2/5. Four Versions. Same Ceiling.

13 min

A controlled cross-version experiment reveals what pipeline improvements actually fix — and what they can't.

2026-03-25

Same Code. Same Models. 15 Versions Later.

11 min

A controlled re-run of our blind audit reveals what pipeline improvements actually matter.

2026-03-24

278 Bugs. 15 Modules. One Thesis.

10 min

What happens when you turn your own multi-model review pipeline against itself.

2026-03-21

We Shipped All 9.

12 min

From research paper to production code — a multi-model review pipeline retrospective.

2026-03-18

48 Research Sessions. 12 Modules. 3 Wrong Labels.

14 min

Reverse-engineering a codebase into theory, then cross-validating every design decision against the literature.

2026-03-14

Anatomy of a Multi-Model Review Pipeline

12 min

The math, systems theory, and cognitive science behind Sweep.

2026-03-14

Every Model Signed Off. Sweep Found 5 P1s.

9 min

What happens when you stop asking one model to review its own work.

2026-03-13

← Back

Blog

13 Phases. 63 Decisions. 138 Bugs in the Spec.

An adversarial code review protocol — designed by one model, scrutinized by every model that didn't build it.

2026-04-0411 min

2/5. Four Versions. Same Ceiling.

A controlled cross-version experiment reveals what pipeline improvements actually fix — and what they can't.

2026-03-2513 min

Same Code. Same Models. 15 Versions Later.

A controlled re-run of our blind audit reveals what pipeline improvements actually matter.

2026-03-2411 min

278 Bugs. 15 Modules. One Thesis.

What happens when you turn your own multi-model review pipeline against itself.

2026-03-2110 min

We Shipped All 9.

From research paper to production code — a multi-model review pipeline retrospective.

2026-03-1812 min

48 Research Sessions. 12 Modules. 3 Wrong Labels.

Reverse-engineering a codebase into theory, then cross-validating every design decision against the literature.

2026-03-1414 min

Anatomy of a Multi-Model Review Pipeline

The math, systems theory, and cognitive science behind Sweep.

2026-03-1412 min

Every Model Signed Off. Sweep Found 5 P1s.

What happens when you stop asking one model to review its own work.

2026-03-139 min