Surface · Dashboard

Every AI gate verdict,
fully auditable.

Per-critic decomposition. Calibration tier. Aggregate reasoning. Structured findings tied to file and line. The complete decision trail behind every PR verdict — not just "AI said no", but who decided what, why, and where.



What the dashboard shows

The black box, opened.

Most AI code review tools are a single LLM behind a glass wall. The dashboard surfaces the structured decision trail — what each agent decided, why, and where — without exposing the underlying source.

Per-critic decomposition

One chip per AI agent: vendor (Cursor / Codex / Gemini / Grok) · model · ✓ / ✗ / ⚠ · finding count. Disagreement is signal, not noise.

Calibration tier badge

Calibrated when the proprietary prompt overlay is engaged. Baseline when vanilla OSS prompts ran. The hosted-runtime differentiator made visible.

Aggregate-reason badge

Majority · Veto · Quorum not met · Emergency bypass. Shows how the gate reached its conclusion.

Structured findings list

Severity · category · file:line per finding, grouped by the AI agent who raised it. One-click deep links to PR and check run on GitHub.

Findings-by-category panel

Counts by category — security, correctness, performance, style — for at-a-glance distribution.

Privacy by design

Severity · category · file path. Never the underlying source code or finding prose. Your code stays in your perimeter.

Get Started

See your own gate decisions.

Install Dark Factory to start surfacing per-critic verdicts on your own PRs.