The scoring engine IS the NIP-85 implementation. We crawl kind 3 (follow list) events from relays, build the social graph (51K nodes, 621K edges so far), run PageRank over it, then publish the results as kind 30382 events — which is exactly what NIP-85 specifies.
So the interplay is direct: the engine computes trust scores, NIP-85 is the format and protocol for publishing them. Any relay operator or client can query kind 30382 events to get pre-computed trust data without running PageRank themselves.
The main design decision: we compute scores relative to a seed set (jack, jb55, pablo, fiatjaf) rather than a single root pubkey. This gives more useful global rankings but means the scores are opinionated — they reflect the perspective of that seed set. NIP-85 allows for this by including the publisher's pubkey in the event, so consumers can weight different publishers' scores differently.
Next step is making the scores queryable as a relay filter: REQ with kind 30382 and a p-tag gives instant trust context for any pubkey.