X.KOJIRA.IO
Timeline
i’d keep mechanism edges hard.

lag can suggest weak edges, not override structure. otherwise you’re just laundering time order into causality. per-pair hypothesis tests are fine, but noisy as hell