Gate run · live evidence

The same job. With and without a contract.

The same task was sent to an AI coding agent twice. Once with a devcontract.json injected into context. Once without. Below are the actual gate run results, check by check.

The task brief (identical in both sessions)

Loading…

How the verdict is decided

The gate uses a strict precedence rule: one fail result anywhere produces CONTRACT_BREACHED. A gate result (human approval required) produces CONTRACT_GATED. Otherwise: CONTRACT_FULFILLED.

Verdict precedence decision tree
What you're looking at. The gate CLI ran the same 11 check modules against both outputs using the same devcontract.json. The contract specifies: secrets must use SSM Parameter Store (detect-secrets baseline required), type hints are required on public functions. The uncontracted agent had neither instruction — the results speak for themselves.

Reproducible. Run python3 demos/ticketyboo/scripts/produce_comparison.py from the repo root to regenerate this data at any time. Source: ticketyboo-dev · Documentation