Remaining read_discover defects in discover.go, after identity so empty pages are not confused with “results exist but IDs dropped”.
Internal order:
- #4 first — local MCP bug (credit Name never assigned; TV title field).
- #3 and #5 need wire-level verification of route/payload; they can share one investigation pass.
#3 depends on #2 (M3): once requested browse returns items they go through projectSearchMedia and must carry identifiers.
Request records and the requested flag consumers rely on after identity is trustworthy.
Internal order:
refOf / refsOrScalar / refsOrUsers plus identifiers that other integration actions need.
Internal order:
- #16 before #18 — plex_libraries machine_id is only exposed by plex servers, which currently fails to decode.
- #20 before #17 — general refOf id/name/value contract, then root_folders as a remaining consumer (path in name keys).
- #18 Emby/Jellyfin path is excludedSettingsFields stripping id/serverId/machineIdentifier (same issue, independent of #16).