Honeycomb latency severity gate to PagerDuty

Adds a judgment layer between a latency alert and a page. The agent investigates how bad the regression actually is — magnitude, breadth of affected traffic, error correlation — and routes accordingly: a real incident pages PagerDuty, while a minor blip becomes a backlog GitLab issue. No more 3am pages for a 50ms wobble.

When your latency alerts are too noisy to page on directly but too important to ignore. Best for teams that want severity-aware routing without writing brittle threshold rules.

1. 1A Honeycomb latency alert fires.
2. 2The agent pulls the affected traces and quantifies regression magnitude and affected request share.
3. 3A branch classifies the event as severe or minor based on impact.
4. 4For severe regressions, the agent triggers a PagerDuty incident with the trace summary.
5. 5For minor ones, it opens a low-priority GitLab issue for later review.
6. 6The chosen output carries the investigation context so responders skip the cold start.