Progress in RDC signoff is not a line going down: it is a line going down for honest reasons. Defect counts can fall because defects were fixed, because findings were reclassified without evidence, or because scope was quietly excluded. Reading progress means telling those apart.
What you see | Honest cause | Dishonest lookalike to rule out |
|---|---|---|
Open defects falling | Fixes landing and re-analysis confirming them. | Defects reclassified as acceptable with no new evidence. |
Unclassified falling | Classification sessions working through the report. | Findings closed in bulk without per-finding dispositions. |
Clean blocks rising | Blocks analysed on current RTL with final intent. | Blocks marked clean on stale RTL or draft intent. |
Coverage holes closing | New sequences running and logging their seeds. | Rows marked covered by tests that never moved reset. |
Five honest mistakes produce a clean dashboard over an unclosed design: intent incomplete so the analysis checked nothing, blocks excluded from scope so their findings never appear, everything waived so argument replaced proof, a stale run predating the latest RTL change, and resets tied off in simulation so dynamic checks passed vacuously. The dashboard counters each one only if its metrics are wired to current evidence: intent revision, scope list, waiver records, run revisions, and per-reset activity.
Re-run analysis after every RTL or intent change and say so on the dashboard. Audit scope against the chip hierarchy each week: every block is either clean on current RTL, open with actions, or explicitly out of scope with a recorded reason. And treat sudden improvement as suspicious: a defect count that halves overnight should trigger an audit of dispositions, not a celebration.
The dispositions: which findings were reclassified, what new evidence each reclassification cites, and who approved it. Fixes without RTL changes deserve scepticism.
As not clean: the change-delta row is open for that block until analysis re-runs on current RTL. Stale clean is not clean.