Verification
Verify the comparison
Check that movement since the last assessment is computed from two measurements, and that nothing unevaluated was reported as closed.
The check that matters most
Confirm that not re-assessed appears as its own state, distinct from closed.
A comparison that lacks it is arithmetically capable of reporting a lost permission as progress. If a pack shows a large number of closures alongside a fallen control population, that is exactly the situation the state exists to expose.
Reconcile the counts
| Check | Against |
|---|---|
| Controls failing now | The complete control index |
| New plus reopened plus persistent | Controls failing now |
| Closed | Controls that were evaluated this time and now pass |
| Not re-assessed | Controls that failed before and were not evaluated now |
Closed plus not re-assessed should account for everything that was failing before and is not failing now. If not re-assessed is zero while the control population dropped, ask why.
Check the gating was applied
| If | Expect |
|---|---|
| The scoring constants changed | Score comparison withheld; control movement still reported |
| The mapping version changed | Materiality band movement withheld |
| A different account | The comparison refused outright |
| No previous snapshot | Rendered as a first assessment, worded as such |
| A snapshot that could not be read | The comparison withheld, and the reason stated |
The last two are different facts and the pack words them differently. A pack that presents an unreadable predecessor as a first assessment would be concealing a failure.
The window
The snapshot is retained for 90 days by default. Beyond that, no predecessor is found and the pack correctly renders as a first assessment. That is a real limitation, and it is published rather than hidden.