Buckets are a planning tool, not a filing system
If a bucket cannot change what you brief next month, it is decoration. The test for any taxonomy is simple: when one bucket outperforms, do you know what to make more of?
Buckets like 'general' or 'other' fail that test immediately. Buckets like 'founder POV', 'customer proof', 'product education' and 'cultural moment' pass it.
Let classification be automatic, let review be human
Automated classification gets you eighty percent of the way at zero marginal cost. The remaining twenty percent — the ambiguous posts — should be surfaced for review rather than silently guessed.
Flagging low-confidence classifications keeps the dataset honest, and an honest dataset is the only kind worth putting in front of a client.