What I actually played.

By raw play count the top of the chart is almost entirely Expedition 33 — "Alicia," "Lumière," "Gustave," each spun fifteen-to-twenty-plus times — interleaved with Soft Hair's "Lying Has To Stop" and the Radiohead deep cuts that anchor every list I've ever produced ("All I Need," "Let Down"). Reranking by minutes promotes the longer Radiohead and Soft Hair tracks, and surfaces "Paranoid Android" — six plays, but six minutes each.

The track-level tail is even longer than the artist-level tail: most things played were played exactly once.

The original essay mapped release decades. The export carries no release years, so this version maps something the curated list never could: when the listening happened. Monthly hours swing by an order of magnitude — a 27.8-hour April against a 2.3-hour February — and the rhythm is plainly seasonal and circumstantial rather than steady. The two tallest months (May 2025, April 2026) bracket the window and coincide with the two largest single-artist obsessions.

Hours of music per month · play-by-play export · partial months at each end

Each month belongs to a different artist — and a different track. Aug '25 and Apr '26 are the obsession peaks.

A heatmap of the top ten artists by month makes the regime-cycling literal. Each row is an artist; each cell is one month's listening hours, scaled to the maximum artist-month cell in the grid (Lorien Testard, Aug 2025, 8.2h). The dark vertical streaks at Aug '25 (Lorien Testard) and Sep '25 (Mac DeMarco) and Dec '25 (Miles Davis) and Feb '26 (Ludovico Einaudi, not shown) are the obsession layer in pure visual form: each artist takes a month and then disappears.

Heatmap opacity scaled to the maximum artist-month cell (Lorien Testard, Aug 2025, 8.2h). Hover any cell for hours.

Spotify exports endTime in UTC, not the listener's clock. Converted to America/Los_Angeles the diurnal curve inverts the obvious reading of the raw file: listening climbs steadily through the early morning, jumps into the workday at 8 am, and hits a hard peak at 9 am (13.6 hours) with a near-tie at noon (13.5 hours). The afternoon dips, briefly rises around 4 pm, then tapers through the evening. This is not after-hours listening; it is clinic-hours and lunch-hours listening — a frontloaded day. The raw 7 pm peak you would see in the export is a UTC artifact, preserved below as a diagnostic comparison.

The weekday picture is stranger. By local day of week the listening is decidedly not uniform: Monday leads at 34.3 hours, Friday nearly matches it at 33.2 hours, Sunday and Tuesday round out the heavy days, and Wednesday collapses to one-seventh of Monday's volume (4.6 hours). The mid-week trough is sharp enough to suggest a recurring Wednesday commitment that crowds music out entirely.

Total hours by local weekday across the window · America/Los_Angeles

The streaming log doesn't carry album names. Albums are reconstructed by joining each play to two metadata sources — the saved-track library and the playlist track list — which still only covers ~30% of music minutes (831 events out of 3,126). The remaining 70% is unattributable at album level from this export: notably, almost all of the Expedition 33 score is missing because that record isn't in saved tracks. Within that lens, one album dominates: OK Computer, at 325 minutes, more than the next two combined. Mac DeMarco appears three times over (This Old Dog, Salad Days, Another One), and the back half is a tour of the saved canon — Pink Floyd, Tame Impala, Connan Mockasin, Khruangbin.

The same split shows up sharply in the data. On the left, the top-saved artists ordered by saved-track count — many sit on the saved shelf without earning much actual play. On the right, the artists who dominate the play log without crossing into the library at all. The two columns barely talk to each other.

Beyond music there are 31.5 hours of audiobooks across 197 sessions and 16.9 hours of podcasts. The audiobook shelf is unusually committed — a handful of titles consumed in long stretches — while the broader shelf is a graveyard of good intentions: more than thirty books opened for under a minute apiece, spanning clinical-trial methodology, software engineering, Japanese lessons, and parenting. The podcast time is concentrated almost entirely in one show.

The spoken-word shelf reads like a self-portrait of its own: a media-and-attention book (The Sirens' Call), a healthcare-AI book (A Giant Leap), a thriller, and a history — bracketed by podcasts about company-building and medical AI. Work and curiosity bleed straight into leisure listening.

Going one level deeper: grouping events into sessions (a new session begins after a 30-minute silence) yields 300 inferred sessions, median 4.7 minutes, with 16 sessions over two hours. Same-artist streaks — consecutive events from one artist with no other artist between them — surface the obsession layer in its rawest form. Both views agree: Lorien Testard in August 2025 is the most acute event in the entire window.

Half of the top-ten streaks are Lorien Testard. Obsession isn't even — it's lopsided.

Spotify also ships two further aggregate views in the export. They cover different windows than the rolling play-by-play file and use Spotify-defined counting rules, so they should not be forced to reconcile with the figures earlier in this essay. They do triangulate behavior from another angle: Spotify's own view of itself.

The Sound Capsule files add weekly Spotify-generated highlights — streak markers, milestone notifications, "you stand out" comparisons against a market average. A small sample of what gets flagged:

"Top entity" weeks reveal monomaniacal stretches with surgical precision — these are exactly the streaks the § X tables surfaced from a different angle.

Put the behavioral data beside the curated essay and four distinct listeners emerge from the same account. The first is the constant: Radiohead, present at the top of every ranking — most minutes played, most tracks saved, the album (OK Computer) that outscores all others combined. This is the load-bearing taste, stable across decades and unchanged by mood.

The second is the obsessive. For a few months the entire chart bends around one new thing — here, the Expedition 33 score — played so intensely it ties a twenty-year favorite. The curated top-100 never captured this because curation happens after the obsession cools; the behavioral log catches it mid-fever. Recency doesn't just bias the data, it is the data.

The third is the archivist — the one who saves 2,060 tracks and follows almost no one, who hoards Pink Floyd and Nirvana and deadmau5 and then plays almost none of them. The library is a self-image; the play log is a self.

The fourth is the professional-curiosity listener. The 31.5 hours of audiobooks and 16.9 hours of podcasts are not background filler — they're another self entirely. Acquired and NEJM AI Grand Rounds sit at the top of the podcast stack alongside Hidden Brain and Ground Truths; the audiobook shelf is anchored by The Sirens' Call on attention and Robert Wachter's A Giant Leap on healthcare AI. Music supplies affect and identity; spoken word supplies intellectual continuity. They live in the same account and almost never run at the same time.

The truer reading is not "three listeners" or even "four listeners" but separate storage systems: saved-library, playlists, search log, play log, Wrapped aggregates, audiobook and podcast shelves — each one describes a different behavioral layer, and a faithful portrait can't collapse them. The first essay was a portrait of the image. This one is the portrait that develops when you let the receipts speak: a morning-and-midday listener, low on skips, prone to month-long fixations, anchored by a canon they keep but don't always reach for.

Artists 1–30 by minutes played, with cumulative share. By artist #30 (Hermanos Gutiérrez) the running total crosses 60% — the remaining 675 artists divide the final 40% between them.