The right phrases — which examples to show.
Which phrases you show matters more than how many. Two sets of the same size can play completely differently.
Last page counted phrases. This one chooses them — and the choice matters more. The model treats your examples as the definition of the job, not loose hints, so the phrases you pick quietly decide what it thinks you are asking for.
Show it three easy, in-key major triads and it learns an easy, in-key move — "land on the plain chord tones." Fine, until a tense chord comes along: a G7#9, where the sharpened ninth wants a bluesy, clashing note against the third. The player that only saw sweet triads will land on G B D — pretty, and wrong for the harmony.
The fix is to choose phrases that mark the line where the harmony bends: include one that resolves an altered chord and one clean landing that contrasts with it. Now the model has heard where the tension sits, not just the easy middle — and it phrases the tricky chord right.
Same number of phrases either way. The difference is entirely in the choosing.
Show the player each set of three, then hear how it resolves the same tense chord. Only one set has heard where the harmony bends.
Same count, different phrases.
Both prompts hold three phrases. Swapping which three — not how many — is what flips the tricky resolution from wrong to right.
It's which, not how many.
Three easy phrases teach an easy move, and the model breaks on the first tense chord. Three well-chosen ones — one that resolves an altered chord, one clean landing beside it — show where the harmony bends, and it holds up. Same count, very different phrasing.
Choose phrases that mark the tense moments, not just the obvious middle. A contrasting pair — one each side of the line — is worth more than a pile of look-alikes.