A vanilla-JS tool that runs on top of the Last.fm API to pull useful structure out of a long scrobble history. It lives at /music/discover and has three tabs.
- Forgotten — artists I played heavily once and stopped playing. The cutoff is “high lifetime play count, zero plays in the last six months.” The list is reliably interesting because it surfaces things I liked but moved past, which is a different signal from “things I forgot about” — those usually weren’t liked enough to play many times in the first place.
- Overlooked — artists I listened to in passing (e.g. a single track over a year ago) but never followed up on. Useful when you’re in a low-investment listening mood and want to give something on probation.
- Connections — artists that share a play-history with my top artists at scale, but that I haven’t played myself. Closer to recommendations, but seeded from my own corpus rather than a generic taste graph.
The discovery tool is tiny — a few hundred lines of JavaScript — and it is more useful than every algorithmic recommender I’ve used. The reason is that it operates over my data with criteria I can read in the source. The opacity of mainstream recommenders is what makes them feel slightly hostile; transparent corpus + transparent filter is friendlier even when the filter is dumber.
Cover art is enriched via iTunes Search where Last.fm doesn’t provide it, which is most of the time for older catalogue entries.