In listeners we trust. Not what they remember. What they do.
89,210Plays·3,174Tracks·3.5M5-Second Observations on Dance UK
Counting the Tune-Outs
The audience ends about where it began, yet tune-outs still pile up along the way.
Latest play · sampled every ~5 seconds
Each red step is one tune-out, sampled every few seconds. Together they make the play’s gross total: the raw signal, before the room is removed.
The Room and the Track
One play could be chance. Across a track’s first dozen plays the pattern holds: nearly every tune-out was the room, not the track.
The average track · mean tune-outs per play · first 12 plays · whole catalog
This is not one track: every track’s first 12 plays are averaged across the whole catalog. Take the room away and the count collapses: the average track’s net tune-outs are the room’s. Its own signal, about 20% of the gross, is what remains above or below the room, down as often as up.
The Early-Play Illusion
The first model corrected the entire early-play curve, including daypart effects that did not actually belong to the new track.
Blue traces the era’s correction; the green dots are the measured deviation from each track’s own settled level.Run backwards, from each track’s last plays toward its first, the same measurement shows the two ends of a track’s life moving in opposite directions. That ruled out the correction quietly measuring itself.
After the Score Settles
A settled score is not a frozen one. The meter keeps weighting recent plays, so it can follow real change: sustained movement is the audience changing its mind about a track, not the meter failing to converge.
Score movement after settling · change from the play-12 score · whole catalog
The line is the median change from each track’s play-12 score; the inner band holds the middle half of the catalog, the outer band 80% of it. Small wander inside the band is the meter’s living resolution. A move that leaves it and stays out is the third kind of movement: trends, fashion, a track’s changing standing with its audience. That is the quantity the meter exists to track.
Library Separation
In this reference comparison, the gold library separates more widely while current rotation clusters tighter.
Cross POP spread · settled tracks · P10–P90
Both stations use the same Cross POP scale. P10–P90 removes the outer 20% of tracks, so a few extremes cannot create the gap. Wider spread on its own is not proof of sharper separation, and the split-half test on the Controls page confirms it here: the 80s library repeats substantially more strongly than current dance.
The Sound Itself
Energy read from the audio alone, knowing nothing about listeners, set beside what listeners did.
Sound energy · 777 tracks · research library
Each dot is one track, the line averages five equal-sized energy groups, the rings mark the low, mid and high thirds. The cloud is wide: energy does not rank a track against its peers, and the fastest genres sit low on this axis. Both reads still climb the same way and both flatten at the top, from a measurement that never saw a listener.How it was measured: Essentia-TensorFlow (Python 3.11), msd-musicnn embeddings feeding two valence-arousal heads, emoMusic and DEAM; energy is the mean of their two arousal values. Every file was loudness-normalized to −23 LUFS before inference, so that loudness-war era mastering cannot enter the measurement, and analyzed at full length in 60-second chunks.
Three reads of music response
POP Score: how well a track holds listeners on this station, within its own environment.
Appeal: how well a track holds listeners across its actual plays, with genre, era, familiarity and format fit retained.
Cross POP: the same underlying listener-response signal, calibrated for comparison across stations.
Live data · Dance UK reference station · based on 11.2M five-second observations across every station measured.
Reference station · Dance UK