When a better AI evaluation score should still block a release
Use a runnable evaluation example to separate critical failures, correct abstentions and incomplete runs before trusting an AI release score.
Blog
Insights on AI, machine learning, computer vision, and enterprise technology.
Active Storage libvips checks before deployment
A patched Active Storage bundle can still fail against an older native library. Check package availability, the deployed image and the library Ruby actually loads.
Accepting AI document results after a file is replaced
Test how document replacements, late extraction results and reviewer decisions affect downstream acceptance with a runnable synthetic example.
Verifying a legacy ACH migration in a Rails application
Test legacy ACH identity, processing states, cutover controls and late returns with a synthetic Rails PaymentIntents migration fixture.
After access is revoked, can your AI search still return the document?
Test how access revocation affects prepared AI answers, citations and caches with a runnable synthetic example and explicit authorization limits.
Retrying a timed-out payment request in Rails
A persisted Rails experiment tests uncertain payment responses, concurrent retries, changed parameters and expired idempotency keys.
Rails upgrade checks
A passing request and successful boot can miss an unchanged upsert value or absent streaming headers. Four checks drawn from the 18 September Rails update.