Symptom: Scoreboard reports mismatch on 5% of transactions. All are multi-ID outstanding reads through a bridge.
Check request queues: are interface, ID, stored address and burst context retained without overwriting repeated IDs?
Verify bridge ID remap: are responses returning with expected IDs?
Check accepted R beats against the correct RID queue head, including RLAST and expected beat count.
Compare failing transactions: same ID reused across ports?
Compare monitor/model evidence with the DUT trace before deciding whether the checker, bridge or endpoint is wrong.
Assertions alone proving system correctness.
Scoreboard key missing port/bridge dimension.
Ignoring false fail rate — team starts waiving real bugs.