Wayback archive of old DR playlists (2012-2021, both page layouts) into archive_plays; synthetic rotation lists (thresholds from exact plays vs lists, applied to archived weeks); Airplay Top-20 position vs DR plays

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-15 04:20:42 +00:00
co-authored by Claude Opus 5.5
parent 992cc2b157
commit 650a6a0f38
4 changed files with 391 additions and 2 deletions
+150
View File
@@ -0,0 +1,150 @@
//! DR playlists from the Wayback Machine: archived copies of dr.dk/playlister/{channel}/{date}/…
//! (2012–2021, mostly P3 2013–2017). The old pages list each play with DR's track id, the exact
//! UTC time, title, artists and DR artist id, the same ids as today, so archived plays join the
//! artist pool directly. Kept apart from the live `plays` table (sampled days, not continuous).
//!
//! airplay collect --only archive
use crate::{db::Db, http::Http};
use anyhow::Result;
use chrono::Utc;
use regex::Regex;
use rusqlite::params;
use std::collections::BTreeMap;
const CHANNELS: &[&str] = &["p3", "p4kbh", "p4", "p5", "p6beat", "p7mix", "p8jazz", "p4fyn", "p4aarhus", "p4nord", "p4vest", "p4syd",
"p4sjaelland", "p4trekanten", "p4bornholm", "p4esbjerg", "p4ostjylland", "p4midtvest"];
pub fn create_tables(db: &Db) -> Result<()> {
db.conn.execute_batch(
"CREATE TABLE IF NOT EXISTS archive_plays (
station TEXT NOT NULL, played_at TEXT NOT NULL, track_id TEXT NOT NULL,
artist TEXT NOT NULL, title TEXT NOT NULL, artist_ids TEXT NOT NULL DEFAULT '',
page TEXT NOT NULL,
recorded TEXT, country TEXT, released TEXT, -- 2012-13 pages: 'Indspillet', country, 'UDGIVET: year, label'
PRIMARY KEY (station, played_at, track_id)
);
CREATE INDEX IF NOT EXISTS archive_plays_day ON archive_plays (station, substr(played_at, 1, 10));
CREATE TABLE IF NOT EXISTS archive_pages (
url TEXT PRIMARY KEY, snapshot TEXT NOT NULL, plays INTEGER, fetched_at TEXT NOT NULL
);",
)?;
Ok(())
}
fn unescape(s: &str) -> String {
let s = s.replace("&amp;", "&").replace("&quot;", "\"").replace("&#39;", "'").replace("&lt;", "<").replace("&gt;", ">");
// numeric entities (&#248; …)
let re = Regex::new(r"&#(\d+);").unwrap();
re.replace_all(&s, |c: &regex::Captures| char::from_u32(c[1].parse().unwrap_or(63)).map(String::from).unwrap_or_default()).into_owned()
}
pub struct Play { pub at: String, pub track_id: String, pub title: String, pub artists: String, pub artist_ids: String,
pub recorded: Option<String>, pub country: Option<String>, pub released: Option<String> }
/// The plays on one archived page. Two layouts: from about 2014 each item carries its UTC time
/// (data-fulltime) and track id; the 2012–13 pages show the local clock time (the date comes from
/// the URL, rolling over at midnight) and add recording year, country and release.
pub fn parse(html: &str, page_date: chrono::NaiveDate) -> Vec<Play> {
let item = Regex::new(r#"(?s)<li[^>]*itemtype="http://schema.org/MusicRecording"[^>]*>(.*?)</li>"#).unwrap();
let head = Regex::new(r#"(?s)<li([^>]*)itemtype"#).unwrap();
let full = Regex::new(r#"data-fulltime="(\d{4}-\d{2}-\d{2}) (\d{2}:\d{2}:\d{2})Z""#).unwrap();
let clock = Regex::new(r#"<time[^>]*>(\d{2}):(\d{2})</time>"#).unwrap();
let tid_attr = Regex::new(r#"data-trackId="([^"]+)""#).unwrap();
let tid_link = Regex::new(r#"/musik/titel/[^/"]+/(\d+-\d+-\d+)"#).unwrap();
let title = Regex::new(r#"(?s)<span class="track" itemprop="name">\s*(?:<a[^>]*>)?([^<]+)"#).unwrap();
let names = Regex::new(r#"<span class="name"(?: itemprop="byArtist")?>([^<]+)</span>"#).unwrap();
let ids = Regex::new(r"kunstner/async/(\d+)/").unwrap();
let details = Regex::new(r#"(?s)<div class="trackDetails">(.*?)</div>"#).unwrap();
let span = Regex::new(r"<span>([^<]*)</span>").unwrap();
let tz = chrono_tz::Europe::Copenhagen;
let (mut day, mut last_min) = (page_date, -1i32);
let mut out = Vec::new();
for m in item.captures_iter(html) {
let whole = m.get(0).unwrap().as_str();
let body = &m[1];
let attrs = head.captures(whole).map(|c| c[1].to_string()).unwrap_or_default();
let Some(ti) = title.captures(body) else { continue };
let Some(track_id) = tid_attr.captures(&attrs).or_else(|| tid_link.captures(body)).map(|c| c[1].to_string()) else { continue };
let at = if let Some(t) = full.captures(body) { format!("{}T{}Z", &t[1], &t[2]) } else if let Some(c) = clock.captures(body) {
let (h, mi): (u32, u32) = (c[1].parse().unwrap_or(0), c[2].parse().unwrap_or(0));
let minute = (h * 60 + mi) as i32;
if last_min >= 0 && minute + 6 * 60 < last_min { day = day.succ_opt().unwrap_or(day); }
last_min = minute;
let Some(local) = day.and_hms_opt(h, mi, 0) else { continue };
use chrono::TimeZone;
match tz.from_local_datetime(&local).earliest() { Some(d) => d.with_timezone(&Utc).format("%Y-%m-%dT%H:%M:%SZ").to_string(), None => continue }
} else { continue };
let info = body.find("class=\"trackInfo\"").map(|i| &body[i..]).unwrap_or(body);
let info = info.find("trackDetails").map(|i| &info[..i]).unwrap_or(info);
let artists: Vec<String> = names.captures_iter(info).map(|c| unescape(c[1].trim())).collect();
let artist_ids: Vec<String> = ids.captures_iter(whole).map(|c| c[1].to_string()).collect();
let d: Vec<String> = details.captures(body).map(|c| span.captures_iter(&c[1]).map(|s| unescape(s[1].trim())).collect()).unwrap_or_default();
let get = |p: &str| d.iter().find(|x| x.starts_with(p)).map(|x| x[p.len()..].trim().to_string());
out.push(Play {
at, track_id, title: unescape(ti[1].trim()), artists: artists.join(", "), artist_ids: artist_ids.join(","),
recorded: get("Indspillet:"), released: get("UDGIVET:"),
country: d.iter().find(|x| !x.contains(':') && !x.is_empty()).cloned(),
});
}
out
}
pub fn collect(db: &Db, http: &Http, log: &mut Vec<String>) -> Result<usize> {
create_tables(db)?;
let path = Regex::new(r"/playlister/([a-z0-9]+)/(\d{4})-(\d{1,2})-(\d{1,2})")?;
let mut pages = 0;
for ch in CHANNELS {
// The CDX index is often overloaded (503): retry patiently.
let url = format!("https://web.archive.org/cdx/search/cdx?url=www.dr.dk/playlister/{ch}/*&output=json&fl=timestamp,original&filter=statuscode:200&collapse=urlkey&limit=50000");
let mut cdx = None;
for attempt in 0..8 {
match http.get(&url) {
Ok(Some(c)) if c.trim_start().starts_with('[') => { cdx = Some(c); break; }
Ok(Some(_)) | Ok(None) if attempt >= 2 => break, // answered, but not an index: nothing archived
Ok(_) => {}
Err(e) => { if attempt == 7 { log.push(format!("archive cdx {ch}: {e}")); } }
}
std::thread::sleep(std::time::Duration::from_secs(30));
}
let Some(cdx) = cdx else { continue };
eprintln!("archive {ch}: index read");
let rows: Vec<Vec<String>> = serde_json::from_str(&cdx).unwrap_or_default();
for row in rows.iter().skip(1) {
let (ts, url) = (&row[0], &row[1]);
if !path.is_match(url) { continue; }
let clean = url.replace(":80/", "/").replace("http://", "https://");
let done: i64 = db.conn.query_row("SELECT count(*) FROM archive_pages WHERE url = ?1", [&clean], |r| r.get(0))?;
if done > 0 { continue; }
let html = match http.get(&format!("https://web.archive.org/web/{ts}id_/{url}")) {
Ok(h) => h.unwrap_or_default(),
Err(e) => { log.push(format!("archive {url}: {e}")); continue; }
};
let Some(c) = path.captures(url) else { continue };
let Some(page_date) = chrono::NaiveDate::from_ymd_opt(c[2].parse().unwrap_or(0), c[3].parse().unwrap_or(0), c[4].parse().unwrap_or(0)) else { continue };
let plays = parse(&html, page_date);
let tx = db.conn.unchecked_transaction()?;
for p in &plays {
tx.execute("INSERT OR IGNORE INTO archive_plays VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10)",
params![ch, p.at, p.track_id, p.artists, p.title, p.artist_ids, clean, p.recorded, p.country, p.released])?;
}
tx.execute("INSERT OR REPLACE INTO archive_pages VALUES (?1, ?2, ?3, ?4)", params![clean, ts, plays.len() as i64, Utc::now().to_rfc3339()])?;
tx.commit()?;
pages += 1;
}
}
Ok(pages)
}
/// Archived days per channel and how much of each day is covered (hours with at least one play).
#[allow(dead_code)]
pub fn coverage(db: &Db) -> Result<BTreeMap<(String, String), usize>> {
create_tables(db)?;
let mut out = BTreeMap::new();
for r in db.conn.prepare("SELECT station, substr(played_at, 1, 10), count(DISTINCT substr(played_at, 12, 2)) FROM archive_plays GROUP BY 1, 2")?
.query_map([], |r| Ok((r.get::<_, String>(0)?, r.get::<_, String>(1)?, r.get::<_, i64>(2)?)))? {
let (s, d, h) = r?;
out.insert((s, d), h as usize);
}
Ok(out)
}
+8 -2
View File
@@ -1,6 +1,6 @@
//! airplay: collects what Danish radio stations play, for modelling when artists get played again. //! airplay: collects what Danish radio stations play, for modelling when artists get played again.
//! //!
//! airplay collect [--db data/airplay.db] [--days 7] [--all-bauer] [--only dr|bauer|hitlisten|deezer|discographies|musicbrainz|spotify|playlists|nmf|nmf-artists|nmf-charts|nmf-tracks] [--backfill] //! airplay collect [--db data/airplay.db] [--days 7] [--all-bauer] [--only dr|bauer|hitlisten|deezer|discographies|musicbrainz|spotify|playlists|nmf|nmf-artists|nmf-charts|nmf-tracks|archive] [--backfill]
//! airplay export nmf-denmark [--out data/nmf_denmark.csv] //! airplay export nmf-denmark [--out data/nmf_denmark.csv]
//! airplay stats [--db data/airplay.db] //! airplay stats [--db data/airplay.db]
//! airplay check [--db data/airplay.db] gaps, duplicates, DR cross-check, chart completeness //! airplay check [--db data/airplay.db] gaps, duplicates, DR cross-check, chart completeness
@@ -11,6 +11,7 @@
//! --backfill fetches all history at once (every DR artist page, the whole Hitlisten archive); //! --backfill fetches all history at once (every DR artist page, the whole Hitlisten archive);
//! without it a run takes a bounded slice, which is what the daily timer does. //! without it a run takes a bounded slice, which is what the daily timer does.
mod archive;
mod bauer; mod bauer;
mod check; mod check;
mod db; mod db;
@@ -60,7 +61,7 @@ fn main() -> Result<()> {
rotations::store_all(&db, &dir, args.iter().any(|a| a == "--reparse")) rotations::store_all(&db, &dir, args.iter().any(|a| a == "--reparse"))
} }
_ => { _ => {
eprintln!("usage: airplay collect [--db FILE] [--days 7] [--all-bauer] [--only dr|bauer|hitlisten|deezer|discographies|musicbrainz|spotify|playlists|nmf|nmf-artists|nmf-charts|nmf-tracks] [--backfill] eprintln!("usage: airplay collect [--db FILE] [--days 7] [--all-bauer] [--only dr|bauer|hitlisten|deezer|discographies|musicbrainz|spotify|playlists|nmf|nmf-artists|nmf-charts|nmf-tracks|archive] [--backfill]
//! airplay export nmf-denmark [--out data/nmf_denmark.csv]\n airplay stats [--db FILE]\n airplay check [--db FILE]\n airplay model panel|backtest|rank\n airplay serve [--addr HOST:PORT] [--prefix /airplay]"); //! airplay export nmf-denmark [--out data/nmf_denmark.csv]\n airplay stats [--db FILE]\n airplay check [--db FILE]\n airplay model panel|backtest|rank\n airplay serve [--addr HOST:PORT] [--prefix /airplay]");
std::process::exit(2); std::process::exit(2);
} }
@@ -139,6 +140,11 @@ fn collect(db: &db::Db, args: &[String], db_path: &str) -> Result<()> {
let r = playlists::nmf_artists(db, &sfc_http, if backfill { usize::MAX } else { 800 }, &mut log); let r = playlists::nmf_artists(db, &sfc_http, if backfill { usize::MAX } else { 800 }, &mut log);
step("new music friday per artist (superfridaychart)", r.map(|n| format!("{n} artists looked up")), &mut log); step("new music friday per artist (superfridaychart)", r.map(|n| format!("{n} artists looked up")), &mut log);
} }
if only.as_deref() == Some("archive") {
// Wayback Machine copies of old DR playlists (only on request: about an hour, once).
let r = archive::collect(db, &http::Http::new(1500)?, &mut log);
step("dr playlists from the wayback machine", r.map(|n| format!("{n} pages read")), &mut log);
}
if wants("nmf") || wants("nmf-charts") { if wants("nmf") || wants("nmf-charts") {
let r = playlists::nmf_global_charts(db, &http::Http::new(1000)?, &mut log); let r = playlists::nmf_global_charts(db, &http::Http::new(1000)?, &mut log);
step("new music friday global charts (superfridaychart)", r.map(|n| format!("{n} Fridays read")), &mut log); step("new music friday global charts (superfridaychart)", r.map(|n| format!("{n} Fridays read")), &mut log);
+7
View File
@@ -5,6 +5,7 @@
//! airplay model backtest [--h 1,4,8] [--refit 1] [--out docs/BACKTEST.md] //! airplay model backtest [--h 1,4,8] [--refit 1] [--out docs/BACKTEST.md]
//! airplay model calibrate [--out docs/CALIBRATION.md] rotation thresholds from DR's lists //! airplay model calibrate [--out docs/CALIBRATION.md] rotation thresholds from DR's lists
//! airplay model nmf [--out docs/NMF.md] New Music Friday Denmark vs DR adds //! airplay model nmf [--out docs/NMF.md] New Music Friday Denmark vs DR adds
//! airplay model synth [--out docs/SYNTH.md] synthetic rotation lists (archive + thresholds)
//! airplay model ifpi the 20-year IFPI test bed -> docs/IFPI.md //! airplay model ifpi the 20-year IFPI test bed -> docs/IFPI.md
//! airplay model adds [--h 0,4] [--out docs/BACKTEST-ADDS.md] backtest of the new-add model //! airplay model adds [--h 0,4] [--out docs/BACKTEST-ADDS.md] backtest of the new-add model
//! airplay model rank [--h 4] [--max-h 12] [--top 50] //! airplay model rank [--h 4] [--max-h 12] [--top 50]
@@ -20,6 +21,7 @@ pub mod kalman;
pub mod nmf; pub mod nmf;
pub mod panel; pub mod panel;
pub mod rank; pub mod rank;
pub mod synth;
use crate::db::Db; use crate::db::Db;
use anyhow::Result; use anyhow::Result;
@@ -59,6 +61,11 @@ pub fn run(db: &Db, args: &[String]) -> Result<()> {
print!("{}", report.markdown); print!("{}", report.markdown);
eprintln!("written to {out}"); eprintln!("written to {out}");
} }
Some("synth") => {
let md = synth::run(db, &p)?;
if let Some(out) = arg(args, "--out") { std::fs::write(out, &md)?; }
print!("{md}");
}
Some("nmf") => { Some("nmf") => {
let md = nmf::run(db, &p)?; let md = nmf::run(db, &p)?;
if let Some(out) = arg(args, "--out") { std::fs::write(out, &md)?; } if let Some(out) = arg(args, "--out") { std::fs::write(out, &md)?; }
+226
View File
@@ -0,0 +1,226 @@
//! Synthetic rotation lists, for weeks without a published list (`airplay model synth` -> docs/SYNTH.md).
//!
//! 1. Learn, per channel and category, how many plays a day a listed song gets: exact plays
//! (DR's playlists) against DR's own lists. The list for the week of the plays may be missing;
//! then songs with the same category on the lists before and after stand in for it.
//! 2. Apply the thresholds to archived playlists (Wayback Machine, 2012–2021): per channel and
//! ISO week, pool the archived plays, scale to a full day by the hours covered, and give each
//! song a category. Stored in `synthetic_rotation` with the coverage behind it.
//! 3. The national series: Airplay Top-20 positions (2001–) against DR plays in the year the two
//! overlap, so each position reads as an estimated number of DR plays a week.
use super::panel::{key, Panel, C, CHANNELS};
use crate::db::Db;
use anyhow::Result;
use chrono::{Datelike, Duration, NaiveDate, Weekday};
use std::collections::{BTreeMap, HashMap};
use std::fmt::Write as _;
const CATS: [&str; 3] = ["A", "B", "C"];
/// Song key: primary artist + title without version tags.
fn song(artist: &str, title: &str) -> String {
let a = artist.to_lowercase();
let mut cut = a.len();
for sep in [" [+] ", " feat. ", " feat ", " ft. ", " ft ", " featuring ", " & ", ", ", " x ", " og ", " med "] {
if let Some(i) = a.find(sep) { cut = cut.min(i); }
}
let t = title.split(" (").next().unwrap_or(title).split(" - ").next().unwrap_or(title);
format!("{}|{}", key(&a[..cut]), key(t))
}
/// List channel for a play station (the ten regional P4s share the national P4 list).
fn list_channel(station: &str) -> Option<&'static str> {
match station {
"p3" => Some("p3"), "p5" => Some("p5"), "p6beat" => Some("p6beat"), "p8jazz" => Some("p8jazz"),
s if s.starts_with("p4") => Some("p4"),
_ => None,
}
}
pub fn create_tables(db: &Db) -> Result<()> {
db.conn.execute_batch(
"CREATE TABLE IF NOT EXISTS synthetic_rotation (
channel TEXT NOT NULL, iso_year INTEGER NOT NULL, iso_week INTEGER NOT NULL,
song TEXT NOT NULL, artist TEXT, title TEXT,
plays_per_day REAL NOT NULL, category TEXT NOT NULL, -- A | B | C (estimated)
days INTEGER NOT NULL, hours_covered INTEGER NOT NULL, source TEXT NOT NULL,
PRIMARY KEY (channel, iso_year, iso_week, song)
);
CREATE TABLE IF NOT EXISTS synthetic_thresholds (
channel TEXT NOT NULL, category TEXT NOT NULL, plays_per_day REAL NOT NULL, songs INTEGER, made_at TEXT NOT NULL,
PRIMARY KEY (channel, category)
);",
)?;
Ok(())
}
pub fn run(db: &Db, p: &Panel) -> Result<String> {
create_tables(db)?;
let mut md = String::new();
writeln!(md, "# Synthetic rotation lists\n")?;
// ---- 1. thresholds from exact plays vs lists
// Lists by week (Monday) and channel: song -> category.
let mut lists: BTreeMap<NaiveDate, HashMap<String, HashMap<String, String>>> = BTreeMap::new();
for r in db.conn.prepare("SELECT list_date, channel, category, artist, title FROM rotation_lists WHERE category IN ('A', 'B', 'C')")?
.query_map([], |r| Ok((r.get::<_, String>(0)?, r.get::<_, String>(1)?, r.get::<_, String>(2)?, r.get::<_, String>(3)?, r.get::<_, String>(4)?)))? {
let (d, ch, cat, a, t) = r?;
let Ok(d) = NaiveDate::parse_from_str(&d, "%Y-%m-%d") else { continue };
lists.entry(d + Duration::days(3)).or_default().entry(ch).or_default().insert(song(&a, &t), cat);
}
// Exact plays per week (Monday, Copenhagen time is close enough at week level), station, song.
let mut plays: BTreeMap<NaiveDate, HashMap<String, HashMap<String, f64>>> = BTreeMap::new();
let mut days_seen: BTreeMap<NaiveDate, std::collections::BTreeSet<String>> = BTreeMap::new();
for r in db.conn.prepare("SELECT station, played_at, artist, title FROM plays WHERE source = 'dr'")?
.query_map([], |r| Ok((r.get::<_, String>(0)?, r.get::<_, String>(1)?, r.get::<_, String>(2)?, r.get::<_, String>(3)?)))? {
let (st, at, a, t) = r?;
let Ok(d) = NaiveDate::parse_from_str(&at[..10], "%Y-%m-%d") else { continue };
let monday = d - Duration::days(d.weekday().num_days_from_monday() as i64);
*plays.entry(monday).or_default().entry(st).or_default().entry(song(&a, &t)).or_insert(0.0) += 1.0;
days_seen.entry(monday).or_default().insert(at[..10].to_string());
}
// The list in force in a week: the published one, or songs with the same category before and after.
let list_for = |week: NaiveDate, ch: &str| -> Option<(HashMap<String, String>, &'static str)> {
if let Some(l) = lists.get(&week).and_then(|m| m.get(ch)) { return Some((l.clone(), "published")); }
let before = lists.range(..week).rev().find_map(|(d, m)| m.get(ch).map(|l| (*d, l)))?;
let after = lists.range(week + Duration::days(1)..).find_map(|(d, m)| m.get(ch).map(|l| (*d, l)))?;
if (after.0 - before.0).num_days() > 21 { return None; }
Some((before.1.iter().filter(|(s, c)| after.1.get(*s) == Some(*c)).map(|(s, c)| (s.clone(), c.clone())).collect(), "bridged"))
};
let mut per_cat: HashMap<(String, String), Vec<f64>> = HashMap::new(); // (list channel, cat) -> plays/day
let mut unlisted: HashMap<String, Vec<f64>> = HashMap::new();
let mut used = Vec::new();
for (week, by_station) in &plays {
let days = days_seen.get(week).map(|s| s.len()).unwrap_or(0) as f64;
if days < 5.0 { continue; } // a week mostly covered
for (station, songs) in by_station {
let Some(lc) = list_channel(station) else { continue };
let Some((list, how)) = list_for(*week, lc) else { continue };
used.push(format!("{week} {station} ({how})"));
for (s, n) in songs {
let ppd = n / days;
match list.get(s) { Some(cat) => per_cat.entry((lc.to_string(), cat.clone())).or_default().push(ppd), None => unlisted.entry(lc.to_string()).or_default().push(ppd) }
}
}
}
writeln!(md, "## 1. Plays per day by category (exact plays vs DR's lists)\n")?;
writeln!(md, "Weeks used: {}.\n", if used.is_empty() { "none yet".to_string() } else { used.iter().take(6).cloned().collect::<Vec<_>>().join(", ") + if used.len() > 6 { " …" } else { "" } })?;
writeln!(md, "| Channel | A median | B median | C median | Not listed, 90th pct | Threshold A | B | C |\n|---|---|---|---|---|---|---|---|")?;
let med = |v: &mut Vec<f64>| { if v.is_empty() { return f64::NAN; } v.sort_by(f64::total_cmp); v[v.len() / 2] };
let pct = |v: &mut Vec<f64>, q: f64| { if v.is_empty() { return f64::NAN; } v.sort_by(f64::total_cmp); v[((v.len() - 1) as f64 * q) as usize] };
let mut thresholds: HashMap<String, [f64; 3]> = HashMap::new();
let made = chrono::Utc::now().to_rfc3339();
for lc in ["p3", "p4", "p5", "p6beat", "p8jazz"] {
let mut m = [0.0; 3];
for (k, c) in CATS.iter().enumerate() { m[k] = med(per_cat.entry((lc.into(), c.to_string())).or_default()); }
let nl = pct(unlisted.entry(lc.into()).or_default(), 0.9);
// Category thresholds: halfway (geometric) between neighbouring medians; C's floor halfway to the unlisted 90th pct.
let gm = |a: f64, b: f64| (a * b).sqrt();
// The list floor (C): the plays/day that best separates listed from unlisted songs (F1).
let listed: Vec<f64> = CATS.iter().flat_map(|c| per_cat.get(&(lc.to_string(), c.to_string())).cloned().unwrap_or_default()).collect();
let others = unlisted.get(lc).cloned().unwrap_or_default();
let mut floor = (m[2] / 2.0, 0.0);
for k in 2..=40 {
let t = k as f64 / 10.0;
let tp = listed.iter().filter(|x| **x >= t).count() as f64;
let fp = others.iter().filter(|x| **x >= t).count() as f64;
let f1 = 2.0 * tp / (2.0 * tp + fp + (listed.len() as f64 - tp));
if f1 > floor.1 { floor = (t, f1); }
}
let th = [gm(m[0], m[1]), gm(m[1], m[2]).max(floor.0), floor.0];
if m.iter().all(|x| x.is_finite()) {
thresholds.insert(lc.to_string(), th);
for (k, c) in CATS.iter().enumerate() {
db.conn.execute("INSERT OR REPLACE INTO synthetic_thresholds VALUES (?1, ?2, ?3, ?4, ?5)",
rusqlite::params![lc, c, th[k], per_cat.get(&(lc.to_string(), c.to_string())).map(|v| v.len() as i64), made])?;
}
}
writeln!(md, "| {lc} | {:.1} | {:.1} | {:.1} | {:.1} | {} |", m[0], m[1], m[2], nl,
thresholds.get(lc).map(|t| format!("{:.1} | {:.1} | {:.1}", t[0], t[1], t[2])).unwrap_or("— | — | —".into()))?;
}
// In-sample check: how well the thresholds reproduce the list in the weeks used.
let mut hit = [[0usize; 4]; 4];
let cat_of = |th: &[f64; 3], ppd: f64| if ppd >= th[0] { 0 } else if ppd >= th[1] { 1 } else if ppd >= th[2] { 2 } else { 3 };
for ((lc, c), v) in &per_cat { if let Some(th) = thresholds.get(lc) { let t = CATS.iter().position(|x| x == c).unwrap(); for ppd in v { hit[t][cat_of(th, *ppd)] += 1; } } }
for (lc, v) in &unlisted { if let Some(th) = thresholds.get(lc) { for ppd in v { hit[3][cat_of(th, *ppd)] += 1; } } }
let listed_total: usize = (0..3).map(|t| hit[t].iter().sum::<usize>()).sum();
let listed_found: usize = (0..3).map(|t| hit[t][..3].iter().sum::<usize>()).sum();
let false_pos: usize = hit[3][..3].iter().sum();
writeln!(md, "\nIn-sample check (song-weeks): listed songs found {listed_found} of {listed_total}; songs wrongly put on the list {false_pos}; exact category {} of {listed_total}. Thresholds are refitted as exact plays and lists for the same weeks accumulate.\n",
(0..3).map(|t| hit[t][t]).sum::<usize>())?;
// ---- 2. apply to the archive
let has_archive: i64 = db.conn.query_row("SELECT count(*) FROM sqlite_master WHERE name = 'archive_plays'", [], |r| r.get(0))?;
writeln!(md, "## 2. Archived playlists → synthetic lists\n")?;
if has_archive == 0 || thresholds.is_empty() {
writeln!(md, "Nothing to apply yet (archive or thresholds missing).\n")?;
} else {
// per (station, iso week): hours covered per day, plays per song
struct Wk { hours: HashMap<String, std::collections::BTreeSet<String>>, songs: HashMap<String, (f64, String, String)> }
let mut weeks: BTreeMap<(String, i32, u32), Wk> = BTreeMap::new();
for r in db.conn.prepare("SELECT station, played_at, artist, title FROM archive_plays")?
.query_map([], |r| Ok((r.get::<_, String>(0)?, r.get::<_, String>(1)?, r.get::<_, String>(2)?, r.get::<_, String>(3)?)))? {
let (st, at, a, t) = r?;
let Ok(d) = NaiveDate::parse_from_str(&at[..10], "%Y-%m-%d") else { continue };
let w = weeks.entry((st, d.iso_week().year(), d.iso_week().week())).or_insert_with(|| Wk { hours: HashMap::new(), songs: HashMap::new() });
w.hours.entry(at[..10].to_string()).or_default().insert(at[11..13].to_string());
let e = w.songs.entry(song(&a, &t)).or_insert((0.0, a.clone(), t.clone()));
e.0 += 1.0;
}
let tx = db.conn.unchecked_transaction()?;
tx.execute("DELETE FROM synthetic_rotation WHERE source = 'wayback'", [])?;
let mut summary: BTreeMap<(String, i32), (usize, usize)> = BTreeMap::new(); // (station, year) -> (weeks, songs listed)
for ((st, y, w), wk) in &weeks {
let Some(lc) = list_channel(st) else { continue };
let Some(th) = thresholds.get(lc) else { continue };
// Only daytime-complete days count fully; scale plays by covered hours / 24.
let hours: usize = wk.hours.values().map(|h| h.len()).sum();
let days = wk.hours.values().filter(|h| h.len() >= 12).count();
if hours < 36 { continue; } // less than a day and a half: too thin to tell a C song from chance
let day_equiv = hours as f64 / 24.0;
let mut n = 0;
for (s, (cnt, a, t)) in &wk.songs {
if *cnt < 3.0 { continue; } // a song needs 3+ plays in the sample to count
let ppd = cnt / day_equiv;
let k = cat_of(th, ppd);
if k == 3 { continue; }
tx.execute("INSERT OR REPLACE INTO synthetic_rotation VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, 'wayback')",
rusqlite::params![lc, y, w, s, a, t, ppd, CATS[k], days as i64, hours as i64])?;
n += 1;
}
let e = summary.entry((st.clone(), *y)).or_insert((0, 0));
e.0 += 1; e.1 += n;
}
tx.commit()?;
writeln!(md, "| Station | Year | Weeks with a synthetic list | Avg songs listed per week |\n|---|---|---|---|")?;
for ((st, y), (wks, n)) in &summary { writeln!(md, "| {st} | {y} | {wks} | {:.0} |", *n as f64 / *wks as f64)?; }
writeln!(md, "\nA week needs at least 36 archived hours and a song at least 3 plays in them; plays are scaled to a full day by the hours covered. A/B is uncertain from a few sampled days (P3: A ≈ 4.7, B ≈ 3.3 plays a day); on P4, A and B get the same plays, so P4 synthetic lists only mean \"on the list\".\n")?;
}
// ---- 3. IFPI Airplay Top-20 position vs DR plays
writeln!(md, "## 3. Airplay Top-20 position → DR plays a week\n")?;
let id_by_name: HashMap<String, Vec<usize>> = p.artists.iter().enumerate().fold(HashMap::new(), |mut m, (i, a)| { m.entry(key(&a.name)).or_default().push(i); m });
let week_of: HashMap<NaiveDate, usize> = p.weeks.iter().enumerate().map(|(i, d)| (*d, i)).collect();
let mut by_pos: BTreeMap<i64, Vec<f64>> = BTreeMap::new();
for r in db.conn.prepare("SELECT year, week, position, artist FROM hitlisten_charts WHERE chart = 'airplay'")?
.query_map([], |r| Ok((r.get::<_, i32>(0)?, r.get::<_, u32>(1)?, r.get::<_, i64>(2)?, r.get::<_, String>(3)?)))? {
let (y, w, pos, a) = r?;
let Some(monday) = NaiveDate::from_isoywd_opt(y, w, Weekday::Mon) else { continue };
let Some(&wi) = week_of.get(&monday) else { continue };
let primary = song(&a, "x");
let primary = primary.split('|').next().unwrap_or("");
let Some(ids) = id_by_name.get(primary) else { continue };
let plays: f64 = ids.iter().map(|&i| (0..C).map(|c| p.y(i, c, wi)).sum::<f64>()).fold(0.0, f64::max);
by_pos.entry((pos - 1) / 5).or_default().push(plays);
}
writeln!(md, "Artist's DR plays (all {} channels) in the chart week, for Airplay Top-20 entries {}–{}:\n", C, p.weeks[0], p.weeks[p.w() - 1])?;
writeln!(md, "| Positions | Entries | Median DR plays | 25th–75th pct |\n|---|---|---|---|")?;
for (b, mut v) in by_pos {
let n = v.len();
writeln!(md, "| {}–{} | {n} | {:.0} | {:.0}–{:.0} |", b * 5 + 1, b * 5 + 5, med(&mut v.clone()), pct(&mut v, 0.25), pct(&mut v.clone(), 0.75))?;
}
writeln!(md, "\nSo the 25-year Airplay Top-20 can stand in for \"in heavy rotation nationally\" (on the order of the plays above), but it only covers the top 20 songs each week; it can't produce B/C lists or anything below.\n")?;
let _ = CHANNELS;
Ok(md)
}