Skip to content

ROADMAP.md

Latest commit

 

History

History
101 lines (87 loc) · 5.1 KB

ROADMAP.md

File metadata and controls

101 lines (87 loc) · 5.1 KB
Raw

Curatarr — Roadmap

Last updated: 2026-08-20. This is the working plan, not a promise — items move as live usage teaches us things. History lives in CHANGELOG.md.

Now (in progress / upcoming)

Reception & archive pillars at full strength

  • Vote mass in the judge — DONE 2026-09, replacing the Trakt idea (Trakt's API went premium-only, de-scoped). The engagement stats Trakt would have sold us were already in our own payloads: TMDB vote_count and OMDb imdbVotes were harvested since forever but never surfaced to the judge — Rating: 7.4/10 read identically at 51 and at 500,000 votes. The verified block now renders Rating: 7.4/10 (24,193 votes). Reception PROSE for movies/shows still fills in via the per-title walkers (movies were at 45/6304 checked when a David Fincher film got pitched with zero reception on file) — the mass is just no longer waiting on them.
  • Wikidata awards — structured Oscars/Grammys/BAFTA data via SPARQL and IMDb/TMDB id mapping, hardening the significance pillar with facts instead of prose distillation.
  • ListenBrainz global popularity — DONE 2026-09: release-group listen/listener counts (MBID-aligned; needs a free account token in LISTENBRAINZ_TOKEN — LB auth-locked these endpoints against scrapers) as deletion-debate and album-dossier evidence, plus derivative-release composition (Live/Compilation/Remix share of on-disk GB) from Lidarr's mirrored MusicBrainz secondaryTypes. True MB relationship edges (cover-of/part-of via recording-rels/work-rels) are de-scoped: the old wording claimed we "already fetch" them, but the code never did — only url-rels for the Deezer bridge. Resolving them would serialize per-recording/work calls through the global 1.1 s MusicBrainz gate, and release-group MBIDs (Lidarr foreignAlbumId) are captured nowhere. Revisit as its own item if album-MBID capture ever lands.

Search stage 2 — LLM facets

  • Extend the enrichment schema with three targeted facet texts (aesthetics / tone-register / relationship-dynamics) so register distinctions ("commercial kitsch" vs. "transgressive art", otome trash vs. subversion) become searchable. Stage 1 (theme facets as individual vectors) is live; stage 2 runs only where stage-1 measurement says the register gap is real.

Remaining external-eval half-fixes

  • Memory decay clock semantics, the dormant loved-recommendation detector, a proper Chroma backup job (one manual snapshot exists), and the drop-confirmation gate. Each needs verification against the eval report before touching code.

Next

Discovery features on the taste-vector stack (SoulSync-inspired)

  • Decade time machine — "my taste, but the 1980s": taste-similarity partitioned by era, one lane per decade.
  • Consensus recommendations — "because you watched A, B and C": scored from taste-vector distances, the curator narrates the connection.
  • Seasonal collections — a seasonal hint for the collection designer (October → horror, December → cozy) — one context line, rotates automatically.

Curation reporting

  • Curation report + yearly review expansion: watch stats, discovery follow-up verdicts, taste drift over the year, biggest additions — the "household year in media" the current report only sketches.

Taste engine stage 3

  • Per-facet taste centroids (the multi-centroid machinery exists) for deletion scoring and recommendations.

Later

  • Release watchlist — follow directors/studios/creators; notify the household when something new lands and it fits the taste profile.
  • Enrichment vocabulary expansion — the tag vocabulary is the truth layer of search; register cases keep a running fixture list.
  • German locale as a global UI/curator language setting.
  • Encryption at rest (PIN) — the flow registers a hash today but encrypts nothing yet; needs an architectural pass (server-side vs. client-side key derivation), deliberately parked.
  • Movie-night ballot — household nomination/vote/watch flow (SoulSync-inspired), a project of its own.
  • Tautulli importer — full historical watch attribution for setups that have it.
  • Extend the music enrichment so the Curatarr actually knows about music
  • Per-episode knowledge — TMDB carries per-episode data (titles, overviews, air dates), but everything here is enriched per SERIES today. Watch history is already per episode, so discussions can state the episode FACT ("S2E5, last night") while honestly disclaiming episode- level plot knowledge. Real per-episode enrichment (its own cache layer, its own staleness, ~20x the rows) is a project of its own.

Standing principles

  • Streaming availability is never a curation factor — the library replaces streaming, not the other way around.
  • Every API we already call gets harvested fully before a new one is added.
  • The LLM never invents facts: verified data in, register-honest prose out; thin evidence skips the judge instead of feeding it.
  • Every live miss the owner catches becomes a test fixture. So please tell me about errors while using it.