Raw Deezer answers are kept so matches can be redone during cleaning. Every strong match's release is stored, so the earliest release date can be taken (Deezer often returns reissues). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Airplay Predictor
Collects what Danish radio stations play and models when an artist is likely to be played again. The methods come from finance and insurance: survival and hazard models (bankruptcy, credit default), self-exciting point processes (trading), regime switching (currencies) and credibility theory (insurance pricing).
Stations: DR P3, every regional P4, P5, P6 Beat and P8 Jazz, plus Bauer Media's NOVA, Radio Soft, The Voice, Pop FM and Radio 100 (all of Bauer's Danish web stations with --all-bauer).
DR extras:
-
each artist's 52-week airplay history per channel;
-
the weekly rotation lists (which songs are in rotation category A, B or C).
-
docs/DATA_SOURCES.md: where the data comes from and its limits. The stations only keep about 7 days.
-
docs/MODELS.md: how airplay works and the models to try.
Build and run
Rust 1.80+.
cargo build --release
./target/release/airplay collect # last 7 days, everything it doesn't have yet
./target/release/airplay stats # what's in the database
Data goes to data/airplay.db (SQLite) and data/rotations/*.pdf.
Tables:
plays: one row per play, in UTC, deduplicated.dr_artist_weeks: weekly plays per DR artist and channel.dr_artists: the raw artist pages.stations,runs.
Collect continuously
The sources forget after about a week. deploy/ has a systemd timer that collects daily (a weekly run would have no margin):
cp deploy/airplay-collect.* /etc/systemd/system/
systemctl daemon-reload && systemctl enable --now airplay-collect.timer
Be polite
One request at a time, at least 0.4 s apart, with a User-Agent that links here. Programmes already read are not fetched again.