Strong candidates name the failure mode in their first or second sentence; weak candidates propose a fix for an unnamed problem. Naming it first does three things: it proves you diagnosed rather than guessed, it constrains which fixes are legal, and it tells the interviewer which follow-up to ask, which keeps you on ground you chose.
Failure mode | One-sentence mechanism | Fix family |
|---|---|---|
Metastability escape | Capture flop resolves late and fans different values to downstream logic. | Legal synchroniser: two or three flops, no logic between them. |
Lost event | A fast pulse falls between two slow sampling edges. | Toggle or stretch synchroniser, or handshake. |
Duplicated event | A slow level spans two fast edges and is counted twice. | Edge detect after synchronising, or handshake. |
Torn bus | Several bits change at once and the destination mixes old and new. | Gray code, handshake, FIFO, or qualified quasi-static capture. |
Stale data | Destination captures before the new data has settled. | Hold data until acknowledged; handshake or FIFO discipline. |
The mapping is many-to-one in reverse: one fix family can cover several modes, but one mode never accepts every fix. A two-flop synchroniser answers metastability escape and nothing else. Saying "two flops" to a lost-pulse question advertises that you never named the mode. The symptom-to-cause table in Failure Modes is the drill sheet for this habit.
Take any symptom, a missed interrupt, a double register write, an impossible bus value, a lockup that never reproduces, and force yourself to say the mode before the fix, out loud, in one sentence. Then add the evidence you would cite: structural check for the synchroniser, formal proof for the protocol, constraints for the timing. That sentence pair is a level-3 answer already; the rubric explains why.
Torn bus: several bits changed at once across per-bit synchronisers. Fix with Gray code, a handshake, or a FIFO, not with more synchroniser stages.
It proves diagnosis, it rules out wrong fixes before they are proposed, and it steers the interviewer's follow-ups onto ground you chose.