Skip to content

Session Log: AV content-model audit — data profile completed on a fresh dump

Date: 2026-09-01 Participants: Than Grove (in a live group session with Yuji Shinozaki and Xiaoming Wang), Claude Code Outcome: docs/planning/av-content-model-audit.md now has a complete, real data-profile section. Confirms the same og_membership-vs-field-storage bug found on Sources also affects AV, plus resolves the corrupted-legacy-field open question with real counts.

1. Fresh dump, this time intact

Than re-exported the AV production dump while C2 (Sources) was underway: mandala-prod-av-db_2026-09-01.sql.gz (186MB — much smaller than the corrupted 1.7GB file from earlier, consistent with that one being a bad export rather than a same-size-but-truncated copy). gzip -t passed. Loaded into DDEV's d7_av database (replacing the earlier failed partial import) and profiled.

2. Confirms the Sources bug also affects AV

The headline finding from C2 — field_data_field_og_collection_ref is empty despite being field_sql_storage-backed, with the real collection-membership data living in og_membership instead — also holds for AV: 0 rows in the Field API table, 11,587 memberships in og_membership (essentially all 11,583 audio/video nodes) plus 85 subcollection→collection memberships matching the subcollection count exactly. This was flagged as a prediction worth checking in the original AV audit draft ("worth double-checking against the actual dump once loaded") — now confirmed. The same migration-source fix applies to both sites: read og_membership, not the Field API value table.

3. Clean numbers elsewhere

  • 11,583 audio/video nodes total (7,396 video, 4,187 audio).
  • Required fields are clean: field_pbcore_title 100%, group_content_access 100%, Kaltura entry-id fields >99.7% filled.
  • field_transcript covers 46.4% of nodes — real, worth knowing before scoping any transcript-dependent UI work later.
  • The historical corrupted-field columns (field_extended_cataloging, field_translation_lang_1/2) still hold 2,482 rows each in production, confirming they must be excluded from any future migration — this closes the open question the original code-only pass could only flag, not confirm.

4. One new anomaly found

68 nodes carry a literal, corrupted MISSING_TYPE value in the raw node.type column — not a display artifact, confirmed by direct query. Not investigated further; filed as a new open question in the audit doc (#8) rather than chased down mid-session, since it wasn't part of what C1 was scoped to resolve.

5. Doc/tracking updates

  • docs/planning/av-content-model-audit.md — data-profile section rewritten with real numbers; "establishes"/"does not establish" sections updated; open questions #6 and #7 marked resolved, #8 added.
  • docs/deferred/av-dump-corrupted-no-data-profile.md — marked RESOLVED, with the resolution recorded and the original problem kept as history. One follow-up item (parameterizing load-d7-source.sh) left open.
  • Sprint 2 backlog: C1's caveat removed, now a clean ☑ like C2.

Next-session starting point

  • C3 (Texts, Than) is the only audit left — same template, and now all three other site audits (Images, AV, Sources) have real data-profile validation to match. Worth checking Texts for the same og_membership pattern too, now that it's shown up twice.
  • Workstream B (Images interactive UI) remains the largest unstarted Sprint 2 piece and has no dependency on C.