Proof
Artists only. Live: every artist lookup NotOnQ has handled, through the playground, the drop-in API or the Qloo harness, updating as it happens. Below it, a fixed benchmark run that measures what live traffic can't.
Benchmark run · Thu, 22 Oct 2026 14:35:02 GMT
A fixed sample, so the numbers mean the same thing for everyone. It compares new artists with chart artists, and tests the stand-ins against Qloo's own answers, which live traffic can't. It used 303 Qloo calls; rerun with node --import tsx scripts/proof.mts.
| Sample | Artists | Usable in Qloo | Unknown (G1) | Not in taste graph (G2) | Gap |
|---|---|---|---|---|---|
| Artists with a release in the last 14 days (ListenBrainz fresh releases, single-artist credits), seeded random sample | 100 | 37 | 61 | 2 | 63% |
| Last.fm global top artists chart | 100 | 96 | 3 | 1 | 4% |
- Fill (30 gap artists, open data and stand-ins only): sourced entity 100%, stand-ins 80%, popularity 33.3%, genres 26.7%, description 3.3%, image 0%.
- Backtest (15 artists Qloo knows, each hidden): stand-ins alone recover 37.3% of Qloo's own top 20 "fans also like" (28.7% of the top 10).
- G1: Qloo has no entity (MusicBrainz ID or exact name). G2: Qloo finds the entity but /v2/insights rejects it as a signal.