Progress & loops — reading a re-review
A re-review returns a fresh list of findings. On its own, "4 findings became 5" tells you nothing — did the old four get fixed, or come back? Two things help you tell.
Fix converges; regenerating can churn
A per-finding Fix changes exactly one thing, so a re-review won't raise it again — it converges. Auto-fix all (the whole-model, review-and-regenerate loop behind the dashboard's Advanced toggle) rebuilds whole layers, so it can fix some findings while re-breaking others it fixed a round or two ago — a layer oscillates between two sets of concerns instead of settling.
So when you want a concern closed, prefer Fix (one precise change) over letting the generative loop rebuild the layer again — and Ignore the subjective ones you accept.
The delta summary
Re-run the whole-model review ("Review all layers") and its cross-cutting summary labels what changed since the previous round, so you can read progress rather than a raw count:
- ✓ resolved — flagged last time, gone now;
- ↻ still open — was flagged in the previous round too;
- ✦ new — surfaced this round (often introduced by a regeneration);
- ↺ recurring — flagged in an earlier round, went away, and is back.
↺ recurring is the signal that a generative rebuild is going in circles — that's your cue to stop the loop and Fix the specific spots by hand, or Ignore what you accept.
The reviewer is a fresh, subjective pass each time, and regeneration rebuilds a layer rather than patching one spot. Both together mean a model won't naturally settle to zero findings — which is expected, not a bug. Prefer Fix to reach a stable, closed state.