Problem
Raw events are only retained for 90 days. Any question with a long horizon — year-over-year growth, cohort retention, whether a feature is being adopted over time — cannot be answered at all.
The workaround so far has been to cook a new table for every chart, which does not scale and loses data whenever we get the shape wrong.
What we'll do
- Retain raw events into cooked table(s) so we keep long history, preserving enough of the event properties to support ad hoc analysis rather than a fixed set of columns.
- Store reusable Kusto queries as shared, versioned notes in the
azure-dev-tools repo, so the team can self-serve answers instead of requesting a new table each time.
Done when
- A retained table exists with history beyond 90 days.
- There is a query/notes library in
azure-dev-tools that the team actually uses.
- A new question can be answered without adding a new cooked table.
Problem
Raw events are only retained for 90 days. Any question with a long horizon — year-over-year growth, cohort retention, whether a feature is being adopted over time — cannot be answered at all.
The workaround so far has been to cook a new table for every chart, which does not scale and loses data whenever we get the shape wrong.
What we'll do
azure-dev-toolsrepo, so the team can self-serve answers instead of requesting a new table each time.Done when
azure-dev-toolsthat the team actually uses.