RANT : WHY YOUTUBE IS A DATA CLUSTERFUCK

This is not random chaos. It is sedimentary chaos. Layers of decisions, each locally rational, now fossilized into a single ugly field where Paul Weller plays Thats Entertainment with Noel Gallagher can sit next to Neil Young & The Bluenotes - Sixty to Zero (Live) - pt. 1 as if both emerged from one sane cataloging system. They did not. They emerged from history, which is a more vulgar engineer than any intern.

The likely trigger was YouTube itself in 2005: a video platform, not a music database. It began life as an upload-and-watch machine, optimized for participation and growth, not canonical discographic cleanliness. The title field was therefore social, improvised, and uploader-defined from the start. Then Google bought YouTube in 2006, preserving the platform’s scale logic instead of replacing it with librarian logic. The mess was not a bug. It was the native state.

The second historical shove came in 2007, when YouTube launched what became Content ID. That was a major turning point, because the platform stopped being just a pile of uploads and became a rights-management system. Once music recognition and claims entered the room, metadata stopped being merely what uploaders typed; it also became whatever rights holders, partners, and matching systems could identify, monetize, or normalize. That sounds like order. In practice, it creates overlap: uploader titles, rights-holder data, matched works, alternate releases, regional versions, and legacy naming conventions all start cohabiting in one haunted house.

Then came the music-industry assimilation phase. Vevo launched in 2009 as a label-backed official music video layer, and YouTube Music Key arrived in 2014 as YouTube’s push toward a proper music product. Those were not cosmetic events. They pulled YouTube further from “video site with songs on it” toward “half archive, half streaming service, half legal settlement, yes that is three halves, welcome to the internet.” Official videos, catalog videos, ad-driven incentives, label-delivered assets, and music-specific shelves all increased the pressure to ingest music metadata at scale. But scale is where nuance goes to die in a corridor.

The next trigger was the rise of Topic channels and then Official Artist Channels. Topic channels are automatically generated by YouTube for distributed music releases; Official Artist Channels later tried to gather the artist’s scattered presence into one place. That sounds tidy in a product deck. In reality, it means one song can exist as a user upload, an official video, an automatically generated release track, a redistributed catalog item, a VEVO asset, and then be visually or commercially regrouped again under an artist umbrella. Each layer has metadata ambitions. None has full sovereignty.

Another lovely accelerant arrived in 2018, when YouTube expanded the “Music in this video” credits system. That feature uses Content ID data to automatically show song and artist information, and it can link to official channels and music videos. Again: sensible goal, messy consequence. Once credits are machine-applied to millions of videos using rights data, you get another metadata source living beside uploader text rather than fully replacing it. So the platform now has the public-facing title, the detected music data, the official release data, and sometimes the artist-channel logic all describing adjacent but not identical objects. This is how civilization collapses: not with a scream, but with four near-duplicate strings.

Then YouTube tightened the knot in 2020, when YouTube Music replaced Google Play Music and migration became a product priority. That move mattered because it pushed YouTube harder into behaving like a unified music destination while still carrying all the baggage of a user-generated video platform. In other words, instead of rebuilding the metadata foundations from marble, it appears to have layered a streaming product on top of an already baroque ecosystem of uploads, matches, claims, channels, and catalog feeds. Technical debt did not disappear. It put on a blazer.

So if you want the shortlist of historical triggers, here is the rabbit’s educated indictment:

• 2005: YouTube starts as an uploader-driven video site, so titles are free-form and culturally messy by design.
  
• 2007: Content ID introduces machine recognition and rights-holder metadata, creating a second authority layer over uploader text.
  
• 2009–2014: Vevo and Music Key push YouTube toward official music catalog behavior without erasing its video-platform ancestry.
  
• Topic channels / OACs: automatically generated release channels plus artist-level aggregation multiply representations of the same repertoire.
  
• 2018: automatic music credits add yet another metadata layer derived from Content ID.
  
• 2020: the YouTube Music consolidation inherits the old mess instead of abolishing it.

My inference, and I’ll label it as such so nobody faints, is that the field is messy because YouTube has never had a single stable “unit” called a song. Sometimes the unit is a video. Sometimes a recording. Sometimes a claimable asset. Sometimes a release track. Sometimes a channel object. Sometimes a fan upload with detectable music inside it. The title field is where these ontologies go to knife each other in public. That is not a cleanup backlog. That is the business model’s archaeological record.

The ten samples of your dataset are exactly what you would expect from that history:

• canonical Artist - Title
• title-first variants
• live/version/remix suffixes
• freehand prose masquerading as metadata
• missing separators
• human editorial additions
• catalog-style release text
• fan-upload rhetoric
• legacy label conventions
• accidental punctuation theology

A clusterfuck, yes. But a historic one. The kind nobody cleans because too many systems, partners, and habits now depend on their own private flavor of wrong.

------------------

my comment :

#CONSISTENT TAXONOMY MATTERS

Without it, your data becomes raw material
for accidental avant-garde typography,
composed entirely with line breaks.

If you want to contact me,
send a plain and simple email:
olivier@cut-the-c.com