Release Impact

Every production deploy is assessed against your governed metrics — automatically, 30 minutes, 2 hours and 24 hours after it ships. No dashboard to build, and nothing to remember to check.

What you need

1. Deploys connected (Settings › Connectors) — or send a release_deployed event with a version property.
2. At least one published governed metric measuring a single event (Metrics). A metric that is a ratio of two events is skipped rather than approximated, because a ratio cannot be attributed segment by segment.

What you get

Open Releases. Each release carries the worst verdict any metric produced:

Verdict	Meaning
Regressed	A metric got worse, and a segment (or the whole population) moved by more than noise explains
Improved	The same, in the good direction
No change	Nothing moved beyond noise — and it says how many segments were tested
Waiting for data	Not enough data to judge — too few observations, or too early; it says which, and when the next check runs

Open a release for the full page: the claim in a sentence, a recommended action, and the evidence — what was tested, what contributed, whether there was enough data, and what would change the conclusion. Export it as Markdown or JSON.

How sure is it?

Tangent only calls a release regressed when the finding survives a multiple-comparison correction across every segment scanned. Checking a dozen dimensions guarantees some will look unusual by chance; the correction is what stops that becoming an alert. On a workspace where nothing actually changed, fewer than 1 release in 20 produces a finding.

The comparison is against the same clock window one week earlier — same weekday, same hour — so a Tuesday-morning deploy is not compared with Monday night.

A release is not a randomised experiment: anything else in the same window (a campaign, an outage, a holiday) is inside the comparison. Every finding says so.

Acting on a finding

Acknowledge, action, resolve or dismiss. The state is kept with who changed it and when, and a re-assessment never resets it.

Contract: docs/DECISION-APPS-SPEC.md in the repository.