Public Incident Note
A public incident note is a cleaned writeup of an operational failure or debugging episode. It should preserve the reusable diagnosis and recovery pattern without exposing private infrastructure or security-sensitive details.
An incident can be summarized as:
Suggested Shape
- Context: generic system class or workflow.
- Symptom: what failed from the user’s perspective.
- Impact: what kind of work was blocked or degraded.
- Diagnosis: how the cause was narrowed down.
- Cause: root cause or best current explanation.
- Fix: public-safe resolution.
- Prevention: monitoring, checklist, test, or runbook update.
- Links: related infra, systems, project, or research notes.
- Promotion: whether the incident should remain a log or update an infra runbook.
Public-Safe Examples
- “A GPU training job repeatedly failed after checkpoint restore.”
- “A batch inference pipeline produced partial outputs after interruption.”
- “A shared environment changed and broke a reproducible run.”
Avoid exact hostnames, usernames, private paths, queue names, ports, internal ticket names, and unpublished experiment metrics.
Checks
- Does the note teach a reusable operational pattern?
- Are infrastructure details generalized?
- Does the prevention step connect to a checklist or runbook?
- Is the issue already fixed or clearly marked as unresolved?
- Would publishing the note reveal security posture or private work?
- Should the prevention step update Incident response for shared research machines?