A life, listened to.

The impulse was simple. I know what I listen to — I have preferences, I make playlists, I can recite my top artists. But the more interesting question was whether my behavior had a shape. Not what I played, but when, and how, and with what kind of consistency.

Do I actually have a “morning playlist,” or is that a story I tell myself? Is late-night listening different in kind from afternoon listening, or just in hour? When I come back to an artist, is it curiosity or comfort?

I pulled my full export — every track, every skip, every platform tag — and enriched each row with the hour, weekday, and session index it belonged to. A session was defined as any sequence of plays without a gap longer than thirty minutes. Simple, but sturdy enough to surface patterns.

The plan had been to layer Spotify’s audio features on top — valence, tempo, danceability. Those endpoints were deprecated a few weeks before I started. What remained was temporal and contextual: when I played something, on what, and how longI stayed.

That turned out to be more than enough.

Three patterns held up under every way I tried to break them.

The week has a signature. Weekday mornings and late weekday evenings were the two reliable peaks. Mid-afternoon, between three and five, listening collapsed — the hours I’m apparently too busy to press play. Saturdays looked almost nothing like weekdays: later start, longer sessions, more variance.

Devices tell a second story. Mobile listening dominated commutes and weekends; desktop took over during work hours. Platform share was a cleaner proxy for what I was doing than the music itself.

Repeat behavior is bimodal. Most tracks I played once or twice. A small core — maybe fifty songs — I played dozens of times, almost exclusively during focused work. Comfort listening and discovery listening barely overlap.

Every finding here describes one person — me. The sample size is fifteen months of a single listener, which is enough to see habits but not enough to generalize. With audio features deprecated, mood and energy remain inferred rather than measured; all the claims I make about “focus listening” or “late-night listening” are behavioral proxies, not acoustic ones.

The session heuristic is also a choice. A thirty-minute gap is defensible but not definitive — a shorter threshold would have multiplied the session count and likely muted the weekend pattern.

The analysis ends where a better version would begin. Fifteen months is a snapshot; I wanted continuous data. GDPR exports are one-off and slow — by the time you look at them, they’re already stale.

So I built the thing that should have existed first: a pipeline that pulls my listens automatically, stores them in Postgres, and keeps the table fresh. That project lives next door.