The easy mistake
Dashboards like to show a status for everything. When a source goes quiet, the simplest thing is to keep showing the last value it sent. For a model release, that turns an old report into a claim about now.
What the recorded run does instead
In the fourth recorded observation, the worker had stopped. cpu-2 was still approved, and nothing reported what was loaded. The record says evidence unavailable, with the reason code RUNTIME_UNAVAILABLE, rather than repeating the last report.
Why a reviewer wants the gap
A reviewer reading the record later can see exactly when certainty ended. That is more useful than a clean timeline that hides it, and it keeps the record honest about its own limits.
Missing is not different
Missing evidence does not establish that another release is running. No usable runtime report was available for that observation. The two call for different responses, so the record keeps them apart.
