Drop a zip (or folder) of Protenix/OpenDDE predictions. Complexes are filtered on median ipTM, consensus epitopes are called from the best models, and binders are clustered by the surface they engage — per target and across targets in a common numbering. Nothing is uploaded: the analysis runs in this tab.
Bimodal by design — the pass mark should sit in the valley, not on a shoulder.
One cell per modelled pair: the family it lands in, or the median ipTM if it did not pass. Cells are labelled, so the colour is a shortcut and never the only cue.
The same files the command-line pipeline writes.