Features
Built around one question: what changed?
Everything below exists to shorten the gap between noticing a number move and knowing which change moved it.
Correlation
Attribution you can put in a postmortem
- Commit-range attribution
- Narrows a regression to a range of commits and publishes a confidence figure. Below 0.7 it reports the range as unresolved rather than picking a favourite.
- Flag-aware
- Feature flag flips are first-class events on the same timeline. A regression behind a 5% rollout is found at 5%, not at 100%.
- Multi-signal
- Latency, error rate, saturation and cloud spend are evaluated together. A deploy that halves latency by tripling instance count is reported as both.
Control
Budgets that page a human rarely
- Error budgets per service
- Define the budget once. Plumbline projects burn rate and escalates only when the projection crosses the window, not when a threshold blips.
- Automatic holds
- A release that is projected to exhaust the budget is held and rolled back before the page fires. The action is written back to the release record.
- Quiet by construction
- There is no alert-rule editor because there are no alert rules. This is the whole design, not a missing feature.
Fit
It goes where your deploys already are
- CI adapters
- GitHub Actions, GitLab CI, Buildkite, Argo. The adapter emits a release event; nothing else changes in your pipeline.
- OpenTelemetry native
- Point your existing collector at Plumbline. No proprietary agent, no second instrumentation pass.
- Export everything
- Every release record, attribution and hold is available over an API and as a nightly Parquet drop into your own bucket.
Next