The Mean Is the Wrong Instrument¶
The Pattern¶
Twelve artists in your collection carry a five-star track inside a mean of 3.45 or below. By the promote/hold/stop protocol every one of them is a non-event — no promotion, no expansion, no further acquisition:
| artist | rated | mean | holds a 5★ |
|---|---|---|---|
| Type O Negative | 28 | 3.00 | "Christian Woman" |
| Pixies | 74 | 3.05 | "Is She Weird", "All Over the World" |
| Guru | 25 | 3.12 | "Lifesaver" |
| The Notorious B.I.G. | 19 | 3.21 | "Things Done Changed", "Gimme the Loot" |
| Radiohead | 56 | 3.21 | "High and Dry" |
| MONO | 9 | 3.22 | "Where We Begin" |
| Mazzy Star | 24 | 3.29 | "Fade Into You" |
| Cannibal Ox | 20 | 3.30 | "Atom" |
| Ed Rush & Optical | 39 | 3.33 | "Chubrub" |
| Weezer | 34 | 3.35 | "Undone", "Pork and Beans" |
| Tori Amos | 30 | 3.40 | "Cornflake Girl" |
Ready to Die holds two five-star tracks. Pixies hold two, across 74 ratings averaging 3.05 — the flattest mean of any major artist you own. On the protocol's arithmetic, Biggie and the Pixies are lane artists at best.
And the extreme case is not on that table because its mean is high. U2: 45 rated tracks, mean 3.71, standard deviation 1.04 — the widest spread of any artist in the collection with real sample. Broken out by album it stops being a distribution and becomes a cliff:
| album | rated | mean |
|---|---|---|
| October (1981) | 2 | 4.50 |
| Achtung Baby (1991) | 12 | 4.42 |
| The Joshua Tree (1987) | 2 | 4.00 |
| The Unforgettable Fire (1984) | 2 | 4.00 |
| Zooropa (1993) | 10 | 3.70 |
| Rattle and Hum (1988) | 1 | 3.00 |
| All That You Can't Leave Behind (2000) | 3 | 2.00 |
| How to Dismantle an Atomic Bomb (2004) | 3 | 2.00 |
Everything through 1993 sits between 3.70 and 4.50. Everything from 2000 sits at 2.00. The mean of 3.71 describes no U2 record that exists.
Historical Context¶
The U2 cliff is dateable to a decision the band made in public. Achtung Baby and Zooropa were the Berlin-and-irony period — Hansa Studios in 1990, the band nearly breaking up, and a deliberate repudiation of the earnest widescreen Americana of The Joshua Tree and Rattle and Hum. The ZooTV tour built an entire stage language out of self-mockery. Then Pop underperformed commercially, and from All That You Can't Leave Behind onward the project was explicitly a return to sincerity and stadium uplift — marketed as such, received as such, and rated 2.00 here.
That is the friction thesis again, running through a single band's career with unusual clarity. The U2 that gets four and five stars from you is precisely the U2 that was embarrassed about being U2. When they stopped being embarrassed, your ratings stopped.
The wider methodological point has a longer history. The mean assumes the thing being averaged is homogeneous — that an artist has a quality level and individual tracks are noisy samples of it. That assumption is fine for Burning Spear and wrong for anybody whose career contains a rupture: a change of label, a change of drug, a reinvention, a death in the band. The album era in particular rewarded artists for making periodic sharp turns, which means the mean is systematically the wrong instrument for exactly the music your collection is densest in.
The corpus has already found this once. The ambivalent-zone-self-knowledge
essay observed of Björk that "the mean doesn't capture that, the outlier
does" — an insight the feedback ledger marked for promotion into the taste
profile as a methodology note. This essay is that observation, tested
against every artist in the library rather than one.
The Recordings¶
Pixies are the clearest demonstration that the problem is structural rather than a matter of taste. Overall 3.05 across 74. But the two five-star tracks are both from Bossanova (1990), and the compilation Wave of Mutilation rates 3.47 — 0.42 above the artist mean. The band did not improve on the best-of. The best-of removed the material that was dragging the average, which is the same thing the mean cannot do for itself.
Radiohead is the case where the system already got this right, and it
is worth crediting. verdicts.json deliberately omits Radiohead with an
explicit note: the 2026-07-17 verdict was genre-split — The Bends strong
at 4.25 on new tracks, OK Computer holding, Kid A Mnesia and A Moon
Shaped Pool weak — and forcing that into a single hold state "would
misread downstream as 'needs more ratings' when the artist already has 60
ratings". That is exactly the right call, made once, by hand. The argument
of this essay is that Radiohead is not the exception. It is the normal case,
and eleven other artists need the same treatment.
Type O Negative was closed in the Closed Items table with the note "bimodal — some peaks, mostly misses". Correct, and then filed as a not-promoted artist anyway. Ataxia — John Frusciante's trio — is the same shape at small scale: Automatic Writing 4.20, AW II 2.80, two records, one artist, 1.4 stars apart.
Ed Rush & Optical at 3.33 carries a five star and sits in the neurofunk lane the feedback ledger already flagged as stale — "jungle-lane ratings have moved on, Ed Rush & Optical remain outliers". The outlier status is the finding, not the noise.
What's Missing¶
Nothing here is an acquisition problem, and it would be dishonest to manufacture a shopping list from it. The gap is in the instrument.
The promote/hold/stop protocol asks one question — is this artist good on average — and answers it well for the case it was designed for: a new artist with fewer than ten ratings where a verdict is genuinely needed cheaply. For artists past twenty or thirty ratings it is measuring the wrong thing, and doing so with increasing confidence as the sample grows. The protocol's own logic says a bigger sample makes the mean more trustworthy; for a bimodal artist a bigger sample makes it more precisely wrong.
Three concrete changes follow, all cheap:
- Record standard deviation and maximum alongside the mean in
verdicts.json. Every number in this essay came from three columns the ledger does not currently carry. - Treat any artist with a five-star track as un-closeable by mean alone. Twelve artists currently sit in that position.
- Make album-level verdicts the default for artists past ~20 ratings, as was already done by hand for Radiohead. U2 is not a 3.71 artist; U2 is a 4.42 album and two 2.00 albums, and only the album-level statement is actionable for acquisition.
The one genuine listening recommendation: Pop (1997) is the missing U2 record between the 3.70 of Zooropa and the 2.00 of the 2000s, and you do not own it. It is the album where the irony broke. Whether it rates 3.7 or 2.0 would settle exactly where the cliff falls.
Acquisition Targets¶
- U2 — Pop (1997) — the single most informative missing record in the collection: it sits precisely at the join between the U2 you rate 4.42 and the U2 you rate 2.00.