One source contained a repeated vector-plot sequence in its machine-readable text layer. Left untreated, that artifact would make the extracted record noisier and less useful for review.
The response was intentionally narrow: adjust the normalizer only for long runs of repeated plot markers, regenerate the affected batch, and re-run the extraction, coverage, locator, duplication, and adversarial checks.
This does not say anything about virality. It does say something about the conditions under which a research record should be trusted: corrections need a boundary, an audit trail, and a chance to fail again.
A public learning project should not hide this kind of repair. The work is stronger when the repair is visible.