ingest: bias yt-dlp picks towards a discography's newest releases

Bandcamp label/artist pages list releases newest-first, so give the first
YTDLP_RECENT_COUNT entries a YTDLP_RECENT_BOOST multiplier when picking,
leaning the radio towards fresh music. Flat extraction carries no dates, so
list position is used as the recency proxy rather than a real date lookup.

The boost is gated on the source URL being a bandcamp discography listing:
single /album/ and /track/ pages (whether a URL-file line or one the pick
recursed into) list tracks in track order, not by recency, and non-bandcamp
sources have unverified ordering, so all of those keep a uniform pick.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

ingest/radieo/config.py

@@ -105,6 +105,15 @@ SUBSONIC_ENABLED = bool(
 YTDLP_URLS_FILE = Path(os.environ.get("RADIEO_YTDLP_URLS_FILE", "/config/urls.txt"))
 # How often to reload the URL list and re-expand container URLs, in seconds.
 YTDLP_REFRESH = float(os.environ.get("RADIEO_YTDLP_REFRESH", "300"))
+# Recency boost. Bandcamp label/artist/discography pages list releases
+# newest-first, so the first entries are the freshest. When picking a release
+# from such a listing, the newest RADIEO_YTDLP_RECENT_COUNT entries get this
+# multiplier on their selection weight, biasing the radio towards new music.
+# 1.0 disables the boost (uniform pick); 2.0 makes fresh releases twice as
+# likely per slot. Applied only at the top level of a source (a discography):
+# once drilled into an album, entry order is track order, not recency.
+YTDLP_RECENT_BOOST = float(os.environ.get("RADIEO_YTDLP_RECENT_BOOST", "1.0"))
+YTDLP_RECENT_COUNT = int(os.environ.get("RADIEO_YTDLP_RECENT_COUNT", "5"))
 
 # --- ListenBrainz suggestions source ---
 # Atom recommendations feed. May be an http(s) URL (the real syndication feed)

ingest/radieo/providers/ytdlp.py

@@ -12,6 +12,11 @@ letter, so a numeric part before the first ``:`` can only be a weight. Lines
 without a prefix default to weight ``1``; a weight of ``0`` disables the line.
 Weights bias the *order* in which sources are tried each pick.
 
+Within a discography (a label/artist page), the newest releases can be given a
+selection boost (``YTDLP_RECENT_BOOST``): bandcamp lists releases newest-first,
+so the first ``YTDLP_RECENT_COUNT`` entries are favoured, keeping the radio
+leaning towards fresh music.
+
 The provider only *resolves* references — it emits ``ytdlp`` tracks whose
 locator is the chosen media URL. The actual download happens in the matching
 fetcher (milestone 4). Flat extraction is cached per source URL to avoid
@@ -21,6 +26,7 @@ re-hitting the network on every pick.
 import logging
 import random
 import time
+import urllib.parse
 from pathlib import Path
 
 from .. import config, tagging
@@ -179,7 +185,7 @@ class YtdlpProvider:
         if len(entries) == 1 and entries[0]["url"] == src_url:
             return entries[0]  # leaf: a track resolves to itself
         pool = [e for e in entries if e["url"] not in recent] or entries
-        entry = random.choice(pool)
+        entry = self._choose(src_url, pool, entries)
         url = entry["url"]
         if url == src_url or depth >= self._MAX_EXPAND_DEPTH:
             return entry
@@ -187,6 +193,45 @@ class YtdlpProvider:
         # Fall back to the entry as-is if that turns up nothing usable.
         return self._pick_entry(url, recent, depth + 1) or entry
 
+    @staticmethod
+    def _choose(src_url: str, pool: list[dict], entries: list[dict]) -> dict:
+        """Pick one entry, favouring the newest releases of a discography.
+
+        Bandcamp label/artist/discography pages return releases newest-first, so
+        the first ``YTDLP_RECENT_COUNT`` entries are the freshest; they get
+        ``YTDLP_RECENT_BOOST`` times the weight of the rest, biasing selection
+        towards new music without needing a per-release date lookup (flat
+        extraction carries no dates).
+
+        The boost only applies when ``src_url`` is such a discography listing
+        (see ``_boostable``). For a single ``/album/`` or ``/track/`` source —
+        whether it is a line in the URL file or one the pick recursed into — the
+        entries are tracks in track order, where position says nothing about
+        recency, so the pick stays uniform, as it does when the boost is
+        disabled (``== 1.0``).
+        """
+        boost = config.YTDLP_RECENT_BOOST
+        if boost == 1.0 or not YtdlpProvider._boostable(src_url):
+            return random.choice(pool)
+        newest = {e["url"] for e in entries[: config.YTDLP_RECENT_COUNT]}
+        weights = [boost if e["url"] in newest else 1.0 for e in pool]
+        return random.choices(pool, weights=weights, k=1)[0]
+
+    @staticmethod
+    def _boostable(src_url: str) -> bool:
+        """True when ``src_url`` lists releases newest-first (a discography).
+
+        The recency boost relies on bandcamp ordering a label/artist/music page
+        newest-first. That does not hold for a single ``/album/`` or ``/track/``
+        page (those list tracks in track order) nor for non-bandcamp sources
+        whose ordering we have not verified, so the boost is limited to bandcamp
+        discography URLs.
+        """
+        if not tagging.is_bandcamp({}, src_url):
+            return False
+        path = urllib.parse.urlsplit(src_url).path
+        return "/album/" not in path and "/track/" not in path
+
     def _resolved_tags(self, locator: str):
         """Corrected identity a previous download wrote into the cached file.