Everything runs on this server. No text leaves the machine and there are no per-scan costs.
Copyleaks scale (default) predicts what Copyleaks would say. It is fitted on 244 of your own Copyleaks-scored texts. Use this to answer “will this pass Copyleaks?”.
Independent scale judges how machine-written the text actually reads, using signals from detector research. It disagrees with Copyleaks sharply and on purpose: Copyleaks passes almost all un-humanized output from your pipeline, and this scale does not.
Neither is a verdict. Every result lists its own caveats — read them.