Test the assumptions behind your AI.
Explore Mallace’s campaign pipeline, evidence model, and development roadmap.
A campaign you can inspect
Mallace’s Python pipeline plans a campaign, selects adversarial cases, executes payloads, evaluates responses, and produces findings and reports. The current validated path is simulated execution.
The prototype today
The local command-line interface and FastAPI prototype are implemented. Reports are exported as JSON, HTML, and SARIF. Mock reports explicitly carry their execution mode.
Evidence integrity first
Live request failures now stop execution rather than silently producing mock responses. Invalid or failed live judging also stops evaluation. Unevaluated responses start as unknown. These safeguards are a foundation; full failure accounting and live acceptance remain open.
An operator workspace is coming
The interface on the homepage is a labeled design concept. A production dashboard, authenticated access, durable operations, and a verified live workflow are planned.
From finding to fix
The proposed ForgeLoop bridge would create a reviewed issue, use ForgeLoop’s existing implementation and PR workflow, and retest an authorized preview of the exact change. The connector is not yet implemented.