But the geometric morphometric analysis identified systematic differences between this reconstruction and the two manual ones, differences that could plausibly reflect the divergence between modern French facial morphology and Upper Paleolithic central European facial morphology — a divergence that the algorithm, trained on modern data, had no way to account for.

None of the three faces was wrong.

Each was a legitimate product of a legitimate method, applied carefully and in good faith to the best available evidence.

But they were different faces.

Different enough that a viewer who saw only one of them would have a substantially different impression of what this woman looked like than a viewer who saw only either of the others.

What This Means for Facial Reconstruction as a Practice