Watch · Runtime health
Runtime health
The Runtime health tab (under Watch, on the hosted dashboard at /team/runtime-health) is the fleet view of own-runtime self-audit findings. Each connected desktop runs the self-audit pass on its own process; when a pass fires a finding (a memory-recycle flap, an RSS climb, an endpoint latency regression) and cloud sync is on, the finding streams up to your team event store. This page groups those findings per install so a team lead sees which machine in the fleet is flapping or wasteful right now, not just a team-wide blur.
What it shows
When findings exist, the page renders two count cards and one card per install:
- Installs with findings and Findings (30d): the number of distinct installs that streamed a finding, and the total finding count over the trailing 30-day window.
- One card per install, grouped by the per-machine id (
sourceMachineHash) the sync pusher stamps on every event. Each card shows a short install label (the first characters of the hash), the finding count, and the latest finding's severity and title.
An event that carried no machine hash groups on its own as an unattributed install, never merged into another machine's bucket. Installs are ordered by the most recent finding first.
Applied-fix outcome
A self-audit finding is a proposal that always waits for approval. When one is approved and applied, the apply step measures the flagged signal before and after and records that delta. The install card reads that outcome back verbatim:
- the applied fix measurably improved the flagged signal,
- the applied fix did not improve it (the honest measured result), or
- the fix was measured with no significant change.
An outcome that was not measured, or whose shape does not parse, shows nothing rather than a guess. The tab never fabricates an improvement.
Where the data comes from
Runtime health reads the same per-team event store that Daemon status, Incidents, and Loop Economicsalready read. There is no new ingest route and no new table: the desktop's runtime metric and self-audit proposal and outcome events flow through the existing local-to-cloud pusher, already redacted at write time. The read is scoped to your team at the database layer, so this page can never see another team's rows.
| Event | What it becomes here |
|---|---|
A self-audit proposal (action.proposed, verb runtime_self_audit) | A finding on its install's card (severity, title). |
Its outcome (action.outcome, linked by parent id) | The measured before/after line under the latest finding. |
Empty state
Until a connected desktop's self-audit pass fires a finding and cloud sync is on, the page shows an honest empty state (“No self-audit findings streamed yet”), never a fabricated table. If the team event store cannot be read at load time, the page says so and asks you to try again shortly.