248 of 251 Origin Flags on This Desk’s Spotify Panel Have Confirmed the Classifier Right Since 12 September, Down From 99.2 to 98.8 Per Cent. A 19-Reading Clean Streak Just Broke

248 of 251 origin flags raised on this desk’s Spotify African panel since 12 September have confirmed the registered country right, 98.8 per cent, down from the 99.2 per cent, 244 of 246 this desk reported two days ago. The drop is one artist: CEDDO, filed as Senegalese by a MusicBrainz name search, turned out on a direct read of the Spotify page to be a German-language pop act, and the correction ends a run of 19 straight readings and 127 flags in which not one row needed changing.

Per-date share of flagged rows confirmed correct. Three dips are the series' only three corrections: 21 September, 28 September and 30 October. Source: anomaly-flag-track-record-extended-30-october-2026.

Show the numbers

Collection date	Confirmed correct, %
2026-09-12	100
2026-09-13	100
2026-09-17	100
2026-09-19	100
2026-09-20	100
2026-09-22	100
2026-09-24	100
2026-09-25	100
2026-09-26	100
2026-09-27	100
2026-09-29	100
2026-10-02	100
2026-10-03	100
2026-10-04	100
2026-10-05	100
2026-10-06	100
2026-10-08	100
2026-10-10	100
2026-10-11	100
2026-10-12	100

Source: This desk's own panel-classification-decisions datasets, one per collection date on which the afw-data anomalies command flagged at least one artist, retrieved 30 Oct 2026 · download the data

Sponsored

Three corrections, three different ways the method failed#

CEDDO charts today on Austria’s Spotify daily chart with 6,427 streams, enough to raise two flags at once: “no African chart presence” and “classified by name search,” the compound this desk’s own prior audit named as its riskiest combination. The registered origin, Senegal, traced back to MusicBrainz matching the artist’s name to the Senegalese warrior-caste term and a 1986 Dakar band called Groupe Ceddo, documented by Music in Africa. Spotify ID 42trS2y9sQiZ14tTr2ghvG is neither: it is a German-language pop and rap act with tracks titled “viele probleme” and “letzter tanz,” charting in Austria because Austria and Germany share a language market, not because any African act travelled there. The roster now reads DE, and the artist is dropped from the African panel entirely.

That is a third correction in 251 flags since 12 September, and it is not the same mistake twice. On 21 September, Seyi Vibez was flagged on a Ghana-Nigeria conflict even though his Nigerian origin was already documented; the fix was writing a manual override that had simply never been entered, not a factual error. On 28 September, Nabi had charted nine straight days as a Guinean artist before this desk’s own classification was found to have read his birthplace, Conakry, and missed his working base, Milan; he was moved to Italy and dropped from the panel. CEDDO is a third kind of failure again: a genuine name collision, where an automated match attached the wrong real-world entity to a Spotify ID entirely. A missing override, a birthplace-versus-residence misread, and a name collision are three distinct failure modes inside the same 32-reading window, which argues against one bug repeating and against the method being free of them.

The same 30 October run flagged a fourth artist by the identical “classified by name search” route and found nothing wrong: Houari Dauphin, registered Algerian, charting only on Morocco’s chart with 5,924 streams. He is Houari Siouani, born in Oran on 10 May 1971, a raï singer active since 1988, and raï’s audience runs across the western Maghreb regardless of border, so an Algerian artist on Morocco’s chart is the desk’s documented export pattern, not an error. It is the first time this desk has flagged him, same as CEDDO, and the two first-time flags landed on opposite sides of the same test on the same day.

The counter-reading#

The test is whether three corrections across a 32-reading, 251-flag record is a real error rate or three accidents sitting inside a small, lopsided sample.

All three corrections this desk has ever made sit inside three single dates out of 32. Remove those three dates and the changed total falls from 3 to 0, a complete removal rather than a partial one, because nothing else in 32 readings has ever required a change. That is expected arithmetic, not a finding on its own: a count built from three rows will always go to zero once those three rows are gone. What it does say is that the other 29 readings, and 248 of 251 individual flags, have never once needed a correction, so the method is not drifting broadly, it is occasionally wrong on specific, identifiable rows.

This desk’s own prior piece on this series warned that “a genuinely new misclassification, in an artist this desk has not yet checked, would not be caught by this table at all until it is flagged once and read.” CEDDO is exactly that scenario arriving two days later: a first-time flag, read once, caught. That is the method working as designed, not a gap in it, but it also means the 19-reading clean streak that preceded it measured something narrower than “the classifier is accurate”: it measured that a small, repeating cast of roughly ten names kept confirming the same decisions, because 244 of the 246 confirmed flags before today were repeat checks on artists already adjudicated, not independent tests of new ones. Houari Dauphin and CEDDO were the first two genuinely untested rows this series has logged since late September, and one of the two broke.

Two further limits do not shrink with more readings. This is this desk auditing its own classifier against sources this desk itself chose to read, not an independent audit. And “32 readings” counts only the dates on which the anomalies command flagged at least one artist; dates the check ran clean are omitted rather than counted as zero, so this is a record of every check this desk’s classifier failed to clear on its own, not a record of every time the panel ran.

What this means for anyone trusting an origin tag#

A flag carrying both “no African chart presence” and “classified by name search” together, the exact pairing CEDDO raised, is now this desk’s highest-confidence signal that a row needs a human to open the actual artist page before the origin ships, not just a biography search on the name. The method that failed on CEDDO, matching text to a name rather than confirming the ID, is the same method that works correctly on the large majority of repeat cases like Soolking or Houari Dauphin, which is precisely why it is dangerous: it is right often enough to be trusted on sight, and wrong on exactly the diaspora, shared-word and shared-name cases a label chasing export numbers is most likely to care about. For a desk or a distributor building its own origin panel from MusicBrainz matches, the working rule this record supports is to treat every first-time, name-search-only flag as unverified until someone reads the artist’s own page, because on this record that is precisely where the errors live, not in the names charting for the tenth time.