Which requirement genuinely calls for Kafka rather than a queue table in PostgreSQL?

Why this is the answer

The distinguishing features are retention and independent offsets: several consumers reading the same stream at their own pace, replaying history after a bug, at a volume a single database cannot sustain. Exactly-once handling by one worker pool, a few ordered durable jobs, and request/response style completion are all comfortable for a queue table with SKIP LOCKED.