A prompt engineered for one model doesn't always land the same way in another — these are organized and tested per model, not copy-pasted across all of them.