Release Passport

Inspiration

Database migrations and backend changes can pass conventional CI while silently corrupting realistic legacy data or making rollback impossible. Release Passport tests the release itself—not just the code diff.

What it does

Release Passport creates isolated baseline and candidate environments, seeds identical production-like data, applies the proposed release, and replays the same API and browser journeys against both. It detects behavioural, schema, aggregate, and data-integrity differences, then executes the rollback and compares the recovered state with the original snapshot.

Deterministic checks establish the facts. GPT-5.6 classifies each difference as intentional, suspicious, or destructive; connects evidence to likely user impact; and generates an actionable recovery plan. The result is a traceable ship/block release passport rather than an unsupported AI opinion.

Current status

The working MVP includes real baseline/candidate database execution, a destructive migration fixture, behavioural replay, data invariants, rollback proof, an evidence dashboard, deterministic safety gates, structured GPT-5.6 analysis, and regression tests. The hackathon build is extending this to isolated Supabase PostgreSQL environments, repository ingestion, API replay, and a browser journey.

How we built it

Codex is used to implement and verify the product end to end: architecture, adapters, migrations, safety contracts, tests, dashboard, and documentation. GPT-5.6 is the reasoning layer over bounded execution evidence; it never fabricates evidence or overrides a proven critical safety gate.

Built With

• codex
• fastapi
• openai-gpt-5.6
• playwright
• postgresql
• python
• supabase