Bring-up arguments go wrong when every piece of evidence is treated as equal. Sort what exists into three piles: what the logs prove, what the clean simulation proved, and what nobody has produced yet.
Evidence | What it proves | What it does not prove |
|---|---|---|
Bring-up logs | The hang is real, corner-correlated, and intermittent with retry recovery. | Where the fault is: logs show symptoms, not crossings. |
Clean simulation | The protocol works when every capture resolves cleanly at nominal timing. | Corner behaviour: ideal clocks have no window, no skew, and no derating. |
Structural report | Whether the pulse path has a legal pulse synchroniser or just two flops. | Whether the protocol is correct under all alignments. |
Missing: width analysis | Nothing yet: nobody has compared the pulse width against the slow period at the corner. | This is the gap the interview answer must name. |
Missing: liveness proof | Nothing yet: no formal check that every source event yields exactly one destination event. | Simulation traffic cannot cover every alignment. |
The honest summary sounds like this: the logs prove a corner-dependent intermittent hang; simulation proves the nominal protocol; nothing yet proves the pulse is capturable at the corner or that the protocol is live under all alignments. That sentence is the Assumption and Evidence steps of the framework, and it tells the team exactly what to produce next. The methods behind each pile are taught in Why Simulation Misses CDC and Formal CDC Checks.
The width analysis: pulse width against the slow sampling period at the corner. If the pulse can fall between slow edges, the loss mechanism needs no metastability at all.