What to preserve now
Save the affected URL, time, visible error, browser or network response, deployment log, provider event ID, and the last change before the failure. Use screenshots or redacted logs; never paste secrets, recovery codes, private keys, or live customer records into the intake.
What not to change during the incident
Avoid dependency upgrades, architecture refactors, global permission relaxation, deleting lockfiles, regenerating authentication, or applying several unrelated AI suggestions. Each new variable makes the cause harder to isolate and the rollback less reliable.
The first stabilization target
Restore the smallest business-critical path or a safe known-good deployment, then prove it with one controlled test. Cleanup and feature work wait until customers, leads, payments, or data are no longer actively affected.
Send impact as well as error text
State whether the whole site is down, which user roles are affected, whether revenue or data is involved, and whether a rollback exists. That information matters more than an invented technical diagnosis.