The Anamata Records corpus: 24 waiata, 18 months of indigenous data governance in practice

# The Anamata Records corpus: 24 waiata, 18 months of indigenous data governance in practice ## What this is An honest look at the first 18 months of running Anamata Records — a small Aotearoa music label — inside an infrastructure designed around Māori data sovereignty. Every release is metadata, every metadata decision is visible, and every consent choice is logged. ## The corpus 24 releases. 4 kinds (waiata, mihi, haka, poi). 5 iwi consent gates. 8 consent log entries. 4 governance log entries. 1 cultural review cycle in the append-only audit log. That's the platform talking. Here's what it looks like in practice. ## What "metadata" means here A waiata's metadata includes the obvious things — title, duration, ISRC, UPC. It also includes: - **kinds**: a list of cultural categories (mihi, waiata, haka, poi) - **cultural_flags**: anything that needs care (e.g. `iwi_consent_required`) - **iwi_consent**: a structured object linking the release to the iwi who consented, the scope, and when - **local_contexts**: which Local Contexts Hub labels are attached (TK Attribution, BC Provenance, Caring notice) That last one is unusual. Most music metadata stops at ISRC + duration. Anamata's metadata keeps going — into the cultural accountability layer. ## The data path When someone wants to publish a waiata, the path is: 1. Author submits a draft (artist + branch lead) 2. The release gets a `cultural_review_status` of `pending` 3. Kaitiaki (cultural reviewers) review the metadata and attached iwi consent 4. If approved, the release can move to `scheduled` then `released` 5. The database **blocks** `scheduled`/`released` transitions while review is `pending` — this is enforced by a PostgreSQL trigger, not just application code That last bit matters. We don't trust the application to do the right thing. The database refuses. This is a different posture from most music platforms — Spotify, Apple Music, Bandcamp all rely on application-level gates for cultural review. A bug bypasses it. A DB trigger doesn't. ## What Local Contexts adds Local Contexts is a non-profit that maintains 42 standard labels for indigenous data. Three of them are now attached to Te Tinihanga: - **TK Attribution** — traditional knowledge, attribution required - **BC Provenance** — bi-cultural provenance (Māori + Pākehā origin) - **Caring notice** — handle with care, replace with care The labels are visual on the page (above the fold of `/waiata/te-tinihanga`). They link to Local Contexts canonical descriptions. The labels are also embedded in the file metadata (XMP for PDFs, ID3 for audio) so downstream platforms see them too. ## What "research-grade" means for the corpus This isn't a blog post. The methodology is: - Every number quoted (24 waiata, 8 consent log entries, 4 governance entries) is queryable from the live database via `/dev/tools/audit` - The audit tool exposes the append-only logs in full - Source for every claim is linked This is the difference between a blog and a publication. A blog says "we have 24 waiata." A publication says "we have 24 waiata, here's the query that proves it, and here's the audit trail of every consent decision." ## What's missing Honest gaps: - Only 1 release has Local Contexts labels attached. We're piloting. - 4 of 24 releases still have `cultural_review_status = pending` — these are scheduled for future releases - The cultural review UI is new (built 2026-07-22); before that, reviews happened out-of-band - Te reo Māori translations of the public surfaces are partial (8/36 pages wired) ## Why this matters Most music industry "data governance" is GDPR checkbox work. The data path stops at "we have consent to use this." For indigenous music, consent is the start, not the end. We need to know: who is this release accountable to? What knowledge does it carry? What care does it need? That's what this corpus is for. 24 waiata, every one of them traced to a decision and a reviewer. --- *Published 2026-07-23. Methodology: queries against the live platform database at the time of writing. For the canonical view, run the queries at /dev/tools/audit.*