Change verification
Prove the fix and prove the site still works.
A successful installer exit is only one signal. Security updates need version evidence, customer-workflow testing, visual comparison, transport and edge checks, monitoring, and a tested recovery path.
Before the change
- Confirm the affected and fixed versions.
- Record current versions and configuration.
- Capture application, database, DNS, and critical configuration backups.
- Verify the backup can be read and identify the restore command.
- Define login, forms, checkout, upload, API, media, and admin tests.
- Record representative desktop and mobile screenshots.
- Confirm maintenance, drain, and rollback owners.
- Set a bounded observation window.
Safe rollout sequence
- Canary. Apply the update to a staging copy, one bounded node, or another safe representative target.
- Validate configuration. Run syntax, dependency, schema, and service checks before accepting traffic.
- Exercise real workflows. Test successful and rejected cases, not only the homepage.
- Compare output. Check expected page markers, referenced assets, responsive layouts, console errors, and content hashes after normalizing only proven dynamic values.
- Verify the edge. Confirm TLS, compression, WAF, cache policy, private bypass rules, and direct-origin denial.
- Expand gradually. Roll through remaining nodes while maintaining capacity and a tested rollback path.
- Observe. Watch errors, latency, authentication, queues, integrity, and customer reports through the defined window.
Evidence that matters
Version proof
Installed package, application, plugin, theme, runtime, or image version matches the supported fixed release.
Behavior proof
The vulnerable behavior is unavailable and normal workflows still succeed.
Fleet proof
Eligible nodes report current health, expected configuration, identical deployed hashes where required, and no failed units.
Recovery proof
The backup, rollback artifact, owner, and recovery procedure are known and time bounded.
Rollback triggers
- Login, checkout, forms, uploads, APIs, media playback, or administrative workflows fail.
- Unexpected redirects, certificate errors, origin loops, private caching, or WAF bypass appears.
- Visual output or required assets differ without an approved application change.
- Error rate, latency, resource pressure, failed units, or integrity findings exceed the change threshold.
- Configuration or deployed hashes diverge between equivalent nodes.
Rollback is not the end. Restore service, preserve evidence, reassess exposure, apply a temporary mitigation where safe, and prepare a corrected update.
Close only with evidence
- Change and completion timestamps.
- Version before and after.
- Backup and rollback identifiers.
- Canary result.
- Workflow and visual test result.
- WAF/cache/TLS/origin result.
- Fleet health and parity result.
- Observation-window result.
- Remaining exceptions.
- Evidence hashes.