mediavocab 2.0.0a2

Reference vocabulary and pydantic data model for media cataloguing.

mediavocab

Reference vocabulary and pydantic data model for cataloguing media works: movies, music, books, comics, games, podcasts, audio dramas, radio, sound effects, and procedural ambient streams — all in a single shared schema.

mediavocab is a foundation library. It defines the vocabulary (enums, genre constants) and the structural models (Work, Release, Entity, Credit, Membership, Appearance). Application logic — provider clients, resolvers, playback, UI — lives outside this package.

Install

pip install mediavocab

The only runtime dependency is pydantic>=2. The taxonomy/ and text/ subpackages import nothing beyond the stdlib, so they are safe to vendor in minimal environments.

Quickstart

from mediavocab import (
    Credit, CreditSection, EntityKind, EntityRef, MediaType,
    RelationRole, Release, VariantKind, Work, WorkRelation, WorkRelationKind,
)
from mediavocab.text import score, work_hash

# Each cut is its own Work (spec §3.4); director's cut links via WorkRelation.
theatrical = Work(
    title="Blade Runner", media_type=MediaType.MOVIE,
    year=1982, runtime=117 * 60.0, production_country="US",
    variant_kind=VariantKind.THEATRICAL,
    credits=[Credit(
        entity=EntityRef(name="Ridley Scott", kind=EntityKind.PERSON),
        role="Director", relation_role=RelationRole.DIRECTOR,
        section=CreditSection.PRINCIPAL,
    )],
)
directors = Work(
    title="Blade Runner", media_type=MediaType.MOVIE,
    year=1992, runtime=116 * 60.0, production_country="US",
    variant_kind=VariantKind.DIRECTORS,
    relations=[WorkRelation(kind=WorkRelationKind.DERIVED_FROM, target=theatrical)],
)

# A Release manifests a Work — many formats, mirrors, packages per Work.
bluray = Release(work=theatrical, container="Blu-ray", region="US",
                 uri="file:///library/blade-runner.mkv")

print(work_hash(theatrical))            # stable SHA-256 identity hash
print(score(theatrical, theatrical))    # 1.0 (self-match)

More walked-through examples in examples/ covering albums, band lineups, radio stations, IoT device routing, work comparison, and the pipeline-sentinel NOT_MEDIA / CONTROL flow.

What's in the box

Module	Contents
mediavocab.taxonomy	MediaType (+ PIPELINE_SENTINELS), VariantKind, ReleasePackaging, EntityKind, OrganisationKind, RelationRole, CreditSection, MembershipKind, TemporalState, ReleaseStatus, StreamMode, WorkRelationKind, ReleaseRelationKind, ContentForm, ProgrammeFormat, AccessibilityKind, PlaybackType, plus GENRE_* string constants. Zero deps.
mediavocab.models	Work, Release, Appearance, Chapter, AccessibilityTrack, AvailabilityWindow, LocalizedTitle, WorkRelation, ReleaseRelation, Entity, EntityRef, Membership, Credit, ExternalIds, License, Signals. Pydantic v2.
mediavocab.text	Normalisation, fuzzy matching, work / release comparison and scoring, SHA-256 identity hashes (work_hash / release_hash), merge with MergeStrategy / IdentityConflict, title parser, content classifier, ISO 639 / 3166 / 8601 / ISBN helpers. Stdlib only.
mediavocab.helpers	Classifier predicates (is_not_media, is_device_entity, is_continuous_release), credit lookups (director, author, performers, filmography_of, episodes_of), and release availability / rights predicates (is_available, release_is_open, release_allows_commercial). Non-normative.

Design highlights

• A type earns its place by changing the schema (A1). SOUND_EFFECT, PROCEDURAL_AMBIENT, AUDIO_DRAMA, MUSIC_VIDEO, etc. each catalogue against different external databases or with different runtime tolerances.
• Devices are entities, not works (A3). EntityKind.DEVICE represents physical playback endpoints. The Work is still a RADIO/MOVIE/MUSIC; the device is how the consumer routes playback. A receiver-class device additionally has a Work counterpart for "turn on the radio" invocation.
• Pipeline sentinels never reach a canonical Work (T8). MediaType.GENERIC, NOT_MEDIA, and CONTROL live on the resolver bag and are rejected at Work construction.
• Each cut is its own Work (§3.4). Theatrical, director's, extended, remaster, fanedit — restructurings of the canonical artefact each get a new Work linked by WorkRelation. ReleasePackaging (deluxe / reissue / box-set / bootleg) is independent — that's how an edition ships.
• PlaybackType is derived from MediaType (A6). AUDIO / VIDEO / PAGED / INTERACTIVE routes resolver dispatch by playback intent. Never persisted on Work or Release. Declare playback_type: ClassVar[Set[PlaybackType]] on each provider.
• Genre is a free List[str] with canonical spellings in mediavocab.taxonomy.genre. ASMR, ambient, anime, adult, etc. are genre tags applied across multiple media types — not types of their own (T1). Programme formats (documentary, concert, talk show) live in ProgrammeFormat, not in genres.

See docs/ for full reference and pattern guides.

Traceability — code cites the spec

The formal specification (docs/mediavocab_spec.md, SPEC_VERSION) is prescriptive: it is the source of truth, and the code implements it. Its §2 axioms (A1–A9) and theorems (T1–T9), together with §3–§8, admit and justify every axis, enum value, model field, and operation.

So the two stay bidirectionally traceable, every public enum, model, and operation carries a greppable clause citation in its docstring naming the axiom / theorem / section that admits it, in the form (spec: <clause>) — e.g. (spec: A1, §3.2/§4.1) on MediaType, (spec: A8a, §3.3/§4.2) on ContentForm, (spec: §6.3, A6) on work_hash. Field-level comments cite the §1.5 identity / routing / description family and, for identity-hash inputs, the hash rule (e.g. content_form → A8b enters work_hash §6.3; routing fields → excluded by A6). Validators cite the axiom they enforce (sentinel rejection → T8; one-MediaType-for-life → A4; org_kind warn → §4.5/A9).

Grep the convention with grep -rn "(spec:" mediavocab/. To find which clause governs a symbol, open it; to find the code for a clause, grep the clause id (e.g. grep -rn "A8b" mediavocab/).

Workspace position

mediavocab sits at the bottom of the stack. Every other package in this workspace depends on it:

                          mediavocab
                              ▲
        ┌───────────┬─────────┼─────────┬───────────┐
        │           │         │         │           │
      tutubo   pyfanedit   pymetal   pyo*…       py_bandcamp / nuvem-de-som
        ▲           ▲         ▲                       ▲
        └────────┬──┴─────────┴───────────────────────┘
                 │
              metadatarr  ◄── canonical resolver, ships every provider above
                 ▲
                 │
           media-archivist  ◄── source-DB orchestrator + sidecars + CLI/server

• mediavocab: vocabulary + structural models (this package).
• tutubo, pyfanedit, pymetal, py_bandcamp, nuvem_de_som, radiosoma, tunein, audiobooker: API clients / scrapers. Each emits mediavocab.Work / Release / Entity directly.
• metadatarr: cross-source resolver framework. Bundles every first-party scraper as a hard runtime dep (no extras juggling) and ships ~24 providers under metadatarr.resolve.providers.
• media-archivist: local source-DB indexer / canonicalizer / CLI / web server. Consumes metadatarr's resolver.

Testing

pip install -e ".[test]"
pytest -q

License

Apache 2.0. See LICENSE.