"Render the same line two ways — the clone and the original — and keep the one that reads best. Per line, not once for everything."
Casting Is a Per-Line Decision
Track 3 kept four independent profiles alive: two clones and the two originals they came from. Studio's recast is where that pays off in production. You can render an utterance across casting choices — most usefully, an original voice against its clone — and compare the two takes of the exact same line. Sometimes the clone nails it; sometimes the original is warmer on that particular sentence. Recast lets you decide per line instead of committing to one voice for the whole piece.
Both Takes Are Real and Kept
An A/B here isn't throwaway. Each recast is a real take, cached and tracked in the utterance's lineage. You keep the one that reads best, and the other remains in history — so a later change of mind doesn't cost a fresh synthesis. This is the coverage-matrix idea: for the lines that matter, you can see how each casting choice sounds and choose deliberately, with the comparisons preserved.
Why Not Just Pick One Voice Globally
A global choice is simpler, but it throws away the exact advantage of keeping originals and clones both selectable. A clone might be excellent across a whole script and just slightly off on one emotional line where the original shines. Recast honors that: the default carries most of the piece, and the specific line that wants the other voice gets it — a fine-grained decision that a single global setting can't express.