Hermes Wiki

PostImplementationVerification PIV

Automated checks run immediately after a change to confirm it actually worked, before the change is considered complete — the network engineering term for what DeveloperTools/Testing/SmokeRegression calls a smoke test.

Why we need this / what value this brings

Catches a bad deploy before it's treated as done, rather than discovering the problem later from a user report or a downstream failure.

When to use this

Immediately after every deployment, automatically, as a required gate — not as an optional or manually-triggered step.

How to use or implement this

Define PIV to check outcomes that actually matter (the device/service functions correctly), not just 'did the deploy command exit 0'; and — the key lesson from the scenario — wire its failure to an automatic action (halt, alert, or trigger rollback), not just a status that a human has to notice.

Research questions

  • When PIV fails, what does the automation actually do — halt and alert, or (as in the scenario) fail without triggering any next step, leaving a human to notice and intervene?
  • Is PIV checking the right things — a failure here means either the upgrade genuinely broke something, or PIV itself has a false-positive bug; which was it in this case?
  • At scale (100+ commands × 1000s of devices), PIV output itself becomes a data-volume problem before AI even gets involved — see DesignPatterns/AlgorithmsDataStructures for diffing the raw output first, and AI/ContextWindowManagement-ChunkingStitching for summarizing what's left.

Empty folder — drop notes, links, and findings here as you research.

Hermes Wiki