Delivery flow metrics

This sensor measures how work flows through the project over weeks: how long changes wait, how soon they get a first review, how old the backlog is, how much CI costs. Loops read it to see whether the way work is done is getting slower or more expensive.

What it sees

A time series taken from the code host: pull requests, reviews, issues and CI runs, counted and timed. It reads trends, not single events: one slow review is noise, a median that climbs for weeks is a signal. It sees how long work takes and what it costs, not why, and not whether the work was any good.

Reading the signal

• Within the margin: each metric's recent median is close to its previous value.
• Worse by more than the margin: the metric is flagged. The flag says where the flow slowed, not what caused it.
• Too few events in the window: a median of a handful of changes is not a trend. A wider window gives a reading.

Gaps in the data (a limit on the code host's API, a repository renamed, a label that changed meaning) can look like a change in the flow. Check that the series is complete before acting on a jump.