ranade-oss makes release checks fail loudly before production
This fork is building a staged deployment process that catches bad releases in rehearsal, before they reach the live legal-AI service.
The early rollout was bumpy, with several same-day fixes to its release process. The later work is more settled: reviewed changes, separate staging diagnostics, and checks on the deployment image before it goes out.
- Rehearsal deployments on Fly.io, a cloud hosting platform, give the team a place to test releases before production.
- Staging health checks now fail closed: an unclear or masked error stops the release instead of letting it proceed.
- Automated checks now lint the deployment workflow itself, catching malformed automation before it runs.
- A release record can refresh automatically on trusted in-house pull requests; the write access and trigger choice deserve a security review.
The team also added a test around keeping one tenant's data separate from another's, though it merits scrutiny because it appears to be the sole test for that protection.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?