Data study
Six maintained systems classify rap and they disagree by a factor of 366. The same company sits at both ends of that range. Spotify's internal taxonomy carries hundreds of rap genres built out of what people actually listen to, while the public interface it shipped to developers offered a single word.
That disagreement is not noise, and this piece argues it is the most useful thing in the data. Six systems built for six different purposes, disagreeing in six patterned ways, produce a map of hip-hop's subcultures that nobody has to share our taste to check. The archetypes below are not kinds of music. They are kinds of existence, read off the shape of a scene's presence across the systems.
The spread
| System | Maintained by | Rap genres | What the number actually counts |
|---|---|---|---|
| Every Noise at Once | Spotify, internal | 366 | 366 by word-boundary match across the rap family. The universally quoted 230 is the count of names containing the literal string "hip hop". Built by algorithmic clustering of listening behaviour. |
| RateYourMusic | community | 107 | 107 nodes under Hip Hop, 116 rendered entries, so 9 sit under more than one parent. Built by editor and community proposal, then a vote. |
| MusicBrainz | community | 100 | Roughly 90 to 110 depending entirely on the matching method used. Built by community-maintained flat vocabulary. |
| Wikipedia | editors | 96 | 96 stylistic entries. 280 bullets in total once 149 geographic and 22 ethnic-fusion entries are included. Built by citation-gated editorial consensus. |
| Discogs | marketplace | 25 | About 25 rap-native styles. The genre page surfaces only 14. Built by fixed editorial style list applied by contributors. |
| Spotify Web API | Spotify, public | 1 | One. The seed is hip-hop. There is no rap, no trap, no drill. Now deprecated. Built by undocumented, frozen, then removed. |
A correction
Every write-up of Every Noise at Once repeats the same figure: 230 kinds of hip hop. It appears in the encyclopaedia entry and in every article downstream of it.
We parsed the map. 230 is reproducible to the digit as the number of genre names containing the literal characters "hip hop", and nothing else. Which means it excludes trap. It excludes drill, phonk, grime, crunk and boom bap. Counted properly across the rap family the figure is 366.
The most quoted statistic about rap's diversity undercounts it by 59 per cent, because it was produced by searching for a phrase rather than by counting a category.
This is not a swipe at the map, which is a remarkable object. It is a warning about how a number travels: once "230 kinds of hip hop" was in one sentence, everyone repeated the sentence and nobody re-ran the count.
⚠️ And a live finding worth recording: the map's dynamic endpoints now return 404. The canonical machine-readable taxonomy of rap is a frozen snapshot that nobody maintains, and Spotify's own developer forum points people to it.
The archetypes
Derived from the pattern of a scene's presence and absence across the six systems, not from a view about the music. A scene that appears in the behaviour-built systems but not in the citation-gated one is in a materially different position from one that appears only in a marketplace's vocabulary, and the difference is checkable.
Present in every system that maintains a rap vocabulary.
Example. G-funk, grime, boom bap, trap
The scene is old enough, distinct enough and commercially large enough that behaviour-driven, editor-driven and retail-driven systems all had to name it. Consensus is a lagging indicator: it means the argument is over.
Present in the behaviour and community systems, absent from Wikipedia.
Example. sexy drill, jersey drill
Listeners and taggers are tracking something the citation-gated system cannot admit yet, because Wikipedia requires published sources and the publications have not caught up. This is where the front edge actually is.
Present ONLY on Wikipedia. No data system recognises it.
Example. mumble rap
A critical term applied from outside rather than a scene label used from inside. Systems built from artist and listener behaviour never adopted it because nobody was ever behaving it. It survives because journalism keeps citing journalism.
Present ONLY on Discogs, often with a huge corpus attached.
Example. thug rap, at 23,916 releases
A category that describes how a record was SOLD rather than how it sounded, frozen around the physical-media era because a marketplace's vocabulary optimises for shelving. Larger than Horrorcore and Cloud Rap combined, and invisible everywhere else.
Present everywhere, filed incompatibly.
Example. grime
Wikipedia files it as a DERIVED genre rather than a subgenre, which is an explicit editorial ruling that it is not hip-hop. MusicBrainz lists it flat alongside rap genres. Every Noise puts seven grime variants inside the rap cluster. The question has already been answered three incompatible ways.
Present in the fast community systems only, absent from the slow ones and from retail.
Example. plugg, sigilkore, digicore
Internet-native, timestamped, and legible mainly to the systems that let anyone add a tag. These leave the clearest documentary trail of any scene in history and the least institutional recognition.
Three disagreements worth the whole piece
1. "Mumble rap" exists only on Wikipedia. Not one data system recognises it. That is the signature of a critical label applied from outside rather than a scene label used from inside: the systems built from artist and listener behaviour never adopted it because nobody was ever behaving it. It persists because journalism cites journalism.
2. "Thug rap" exists only on Discogs, with 23,916 releases attached. That is a larger corpus than Horrorcore at 13,041 or Cloud Rap at 8,677, and it is invisible in every other system. Discogs is a marketplace for physical objects, so its vocabulary is frozen around the era of the objects and optimises for how a record was sold rather than how it sounded. An enormous category that only a shop can see.
3. Whether grime is hip-hop has already been answered three incompatible ways. Wikipedia files it under derived genres rather than subgenres, which is an explicit editorial ruling that it is not. MusicBrainz lists it flat alongside rap genres. Every Noise puts seven grime variants inside the rap cluster. Answering it a fourth time from taste would add nothing. Reporting that three maintained systems have ruled differently is the finding.
Method
The Every Noise map was parsed directly for genre names. MusicBrainz came from its API. RateYourMusic's tree was read in a browser. Discogs style facets were read with their release counts. Wikipedia's list was counted from the wikitext rather than from the rendered page. Where a published figure existed we re-derived it, which is how the 230 finding surfaced at all.
This cuts against us as much as anyone. MusicBrainz returns roughly 90 to 110 rap genres depending on how you match, and we report the range rather than picking the flattering end. RateYourMusic renders 116 entries for 107 nodes, because nine sit under more than one parent. Keyword counting undercounts every system in the same direction, which is precisely the trap the famous 230 fell into.
Regional revenue figures were the weakest layer in the research and one widely cited source turned out to be synthetic statistics with no traceable origin, so no economic figures appear in this piece at all. A trade report we wanted is quoted everywhere for numbers its own public landing page does not contain, so we did not repeat them. Figurehead attributions were sourceable for only four of twenty-five scenes, and every one of those four is a recent internet-native scene that left a timestamped trail. ★ Every competitor gets that confidence gradient exactly backwards, writing the 1990s canon with total certainty and the new scenes with hedges, when the documentary evidence runs the other way.
Re-run any count above. They are all reproducible in a browser, and three of the six are one API call. If a number here does not replicate, it is wrong and we want it corrected. The archetypes are a harder target: to break them, find a scene whose presence pattern across the six systems does not fit any of the six shapes.
The measurements are open. Published as CC-BY 4.0 JSON with the method attached.
Sources