Metadata Production create, link and enrich every record

A record moves through defined stages, with human review built in, not bolted on at the end:

1. Classification and normalisation of the incoming data
2. Enrichment, automated where the data is structured, flagged to an editor where it is ambiguous
3. Mandatory editorial review, every record passes at least one human editor before delivery
4. QA validation against the platform's rule set, with critical errors blocked from export
5. Versioning, so every change is tracked
6. Delivery, through Content Delivery, with near-real-time updates within minutes of processing

Each incoming broadcast or title is matched against the content archive automatically. High-confidence matches are linked straight through, and the entry inherits the master record's metadata (cast, director, synopsis, year). Lower-confidence matches are not guessed: they are surfaced as suggestions for an editor to confirm. Matching uses title, episode number, series name, production year and genre context. Once linked, the entry joins the relational content graph (series, season, episode), so the same title keeps one identity no matter how many channels carry it or how each source labels it.

Alongside the core fields, records are enriched with:

• cast and crew, linked down to character level
• multi-length synopses, from very short to long
• a canonical genre taxonomy and thematic tags
• age ratings from official national bodies
• editorial flags such as premiere, live or final episode

Enrichment draws on trusted external sources (IMDb as the primary, alongside others), automatically for structured data and with an editor for anything ambiguous. Special events are handled as first-class objects, with their own Event ID, parent-child structure and links to sport and awards data.

The platform runs 500+ automated QA rules, developed with the editorial team and covering both global standards and per-client contractual requirements. They check completeness (missing titles or episode numbers), format correctness (year and date syntax), consistency (duplicate episodes, numbering gaps), standard compliance (description length) and required metadata (genre tags, age ratings). Critical errors are blocked from delivery until resolved, and every record still passes at least one human editorial review. The principle is simple: automation where it is safe, human oversight where it matters.

Where AI contributes to a field, its work is logged, model, version, confidence and the editor's review decision, as an immutable audit trail. So you can always tell what was produced automatically and what an editor confirmed.

Every object receives a permanent ID, constant across channels, languages and platforms. It is what lets a title be tracked and connected long after the original source identifiers have changed, and it is the foundation the rest of the platform, and products like Common ID, build on.