10.11.2.1 (beta)

Discogs (new; needs a Discogs token):

• Complete curated record labels, chosen with the
  type-ahead chip picker (search, pick a match, it
  becomes a removable chip).
• Complete an owned artist's discography, covering
  artists the MusicBrainz sources cannot (no
  MusicBrainz id), with conservative exact-name artist
  resolution.
• Resolve real id->label names.
• Recognize an owned album by a conservative
  artist-and-title name match as well as by Discogs id,
  so it works on MusicBrainz-tagged libraries.

Diagnose (deeper and wider):

• Deeper analysis compares external ids to tell a real
  misidentification from a coincidental title clash.
• Episode and season gaps get their own diagnosis
  that spots a same-named reboot mis-tagged onto
  a series (the V 1984 versus V 2009 case).
• Album and book gaps are diagnosable.
• The Diagnose button shows on every gap kind.
• Name-plus-year matching prefers an exact-year hit
  over the one-year tolerance.

Reboot handling:

• The series cross-checks skip a same-named reboot
  resolved for a series, by year.
• The library source no longer reports a reboot's later
  seasons as missing episodes of the single-season
  original (it compares against the owned-episode year
  range).

Music and books:

• The OpenLibrary Books source is hardened using
  real author data: resolve the author from an owned
  work, a conservative shortest-name matcher for the
  name search, works listed via the search endpoint
  so they carry a year, and de-duplication of works
  that share a title. Captured fixtures added for the
  new endpoints.

External calls (stay clear of rate-limit bans, scan faster):

• Every external API call now goes through a
  read-through memory cache, so a repeated lookup
  within or across scans does not hit the network again.
• Proactive per-service rate pacing for MusicBrainz
  and Discogs (one request a second).
• A per-service circuit breaker moves the scan past a
  down service, and each source skips a service once
  it has been given up on.
• Stable id-to-name lookups (studio, keyword, and
  label names) are cached for far longer.
• The scan runs its sources concurrently, so a slow
  provider no longer holds up the fast ones, and
  streams results into the report per item, so gaps
  appear as they are found.

Dashboard and UI:

• The Type chooser always shows every media
  domain, even one with zero entries this scan.
• The settings panel no longer reloads the report
  on scan-status polls while it is open.
• The dashboard is authored as separate report,
  settings, CSS, and JS files, concatenated at build
  time.

Housekeeping:

• Shared TextKey.Normalize for title and name
  matching; one ServiceNames constant for the
  HTTP service names; the Discogs provider-id
  constant for the artist id check; explanatory
  comments removed from the build files;
  README, catalog description, and roadmap
  updated for the grown scope.