B2

An individual’s listening history usually comes back as a ranking of top artists and top songs once a year for review. A ranking counts volume and throws away the calendar, which is the only axis on which evolution of taste can be observed. B2 reads the same record as a shape in time where each recurring song sits at the month it peaked, raised or lowered by how bright or bleak its lyrics are. The claim is that the words I keep returning to trace a line across six years and songs played back-to-back mark real adjacencies.

Key question

Does a throughline exist for repeated themes in my listening history?

My extended streaming history was recovered locally at 273,810 play events across 75 months, April 2020 to June 2026, of which 142,297 cleared the thirty-second bar counted as a real listen, a retention rate of 52.0%. Podcasts and audiobooks were partitioned out, overlapping exports de-duplicated, and timestamps converted to local time before any calendar was drawn.

The 618 songs that recurred often enough to qualify were given lyrics from LRCLIB and scored on-device by a language model on two things: how bright or bleak the words are, zero to one, and what themes recur. Those songs carry 45,475 listens, 32.0% of the total, from 265 artists. A song sits horizontally at its peak month and vertically at its brightness. Two songs are joined when they were played back-to-back more than once, and 2,132 such handoffs form the web behind the dots.

The collection sits below the midpoint and stays there. The median song scores 0.35 against a neutral 0.50, and the yearly medians hold between 0.32 and 0.35 in every calendar year from 2020 to 2026 but one. The departure is 2024, at 0.42 across 92 songs, which reverts the following year.

What repeats is a vocabulary. Of 28 themes found in the lyrics, 14 appear in all seven calendar years. The largest, longing, carries 120 songs, or 19.4%, against 12.5% for devotion in second place.

The themes separate sharply on brightness. Intimacy sits at 0.67 and devotion at 0.62, against 0.25 for isolation and 0.27 for obsession, a spread of 42 points. The brightest and bleakest are both present in all seven calendar years.

One artist supplies 80 of the 618 songs and 16.9% of the recurring plays, roughly triple the next at 31. The median song stays in rotation 1,054 days, just under three years.

Selecting a theme from the left lights the songs that carry it and walks through them forward in time, centering each and playing a thirty-second clip. Height is brightness, so the walk rises and sinks as the language changes.

Nothing on the page interprets. A song card reports its brightness and how long it stayed in rotation, the theme names come from the lyrics, and the walk orders by calendar.