Skip to content

Session Log: AV content-model audit (Sprint 2 Workstream C1)

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 written and committed. Structural findings complete and code-verified; the planned data-profile section could not be completed because both local copies of the AV production dump are corrupted.

1. Why C1 first

Following the theme install on dev-0 (see the earlier 2026-09-01 log), Sprint 2's Workstream A is closed and B/C/D are all unblocked. C1 (the AV audit) was picked to start with since it's group-owned and has no single source module to assign to one person — following the sprint doc's own note that it should "take longer to scope than C2/C3," it made sense to start it in this live group session.

2. Correcting a stale prior note

A prior team note (carried in Claude's own project memory) assumed mb_metadata and mb_structure owned custom database tables analogous to Images' shanti_images sidecar table. Static analysis of both modules found no hook_schema() or hook_node_info() in either — AV is just two plain node bundles (audio, video), defined by a Features export sub-component of mediabase (mediabase/features/audio_video). This correction is recorded in the audit doc itself, not just here.

3. Structural findings (code-verified)

  • Full field inventory for both bundles (near-identical, differ only in field_audio/field_video) — PBCore descriptive/technical metadata and cataloging workflow state are all field_collection items, a materially different (and likely lower-risk) starting shape than Images' referenced-node satellites.
  • The Kaltura integration boundary: the node holds only a scalar Kaltura entry ID; mb_kaltura is a customized import/sync layer (the stock contrib import UI is deliberately disabled — "their import mechanism is broken," per its own code comment). Ties into Spike 7 (○ Pending), not resolved by this audit.
  • Transcripts attach to the same node (file field + a separate timing table keyed by node ID), not a separate entity.
  • AV's access model is OG plus two additional custom grant realms (group_access_uva_member, mb_collection_admin) not present in Images' model — flagged so a future OG→D11-Group mapping doesn't silently drop them by copying Images' mapping.

Full detail, field-by-field, in the audit doc.

4. Data profile blocked — dump corrupted

Attempted to load the real AV production dump into DDEV (d7_av database) to validate the structural findings against real rows, the same way the Images audit validated its Paragraphs decision against real agent-reuse numbers. Both local copies failed:

  • ~/Sandbox/Mandala/data/mandala-prod-av-db_2026-06-11.sql.gz (1.7GB) — gzip -t fails (unexpected end of file); confirmed live by attempting the load, which aborted mid-decompression and fed a truncated line into MySQL.
  • ~/Sandbox/Mandala/mandala11/data/mandala-prod-av-db_2026-06-11.sql.gz — same filename but actually a different, much smaller (24.8MB) file, not a real duplicate/backup; also gzip-corrupt.

Cleaned up the ~16GB partial d7_av database created during the failed import before it could cause confusion for anyone else on the shared DDEV instance. Filed av-dump-corrupted-no-data-profile.md, cross-referenced against the existing canonical D7 dev source dump note since both point at the same underlying gap (dump provenance/integrity isn't tracked anywhere).

5. Doc status

docs/planning/av-content-model-audit.md is explicit throughout that its findings are static-code-derived only, not row-verified — every section that would normally cite real numbers (data profile, corrupted-legacy-field check) instead states clearly that verification is pending a fresh dump. Sprint 2 backlog row C1 marked ☑ with a caveat link to the deferred note; the sprint's own acceptance criterion for the audits (real dump, not placeholders) stays unchecked until a working dump is available.

Next-session starting point

  • Obtain and gzip -t-verify a fresh AV production dump (out-of-band, per the deferred note), then re-run the audit's data-profile section.
  • C2 (Sources, Xiaoming) and C3 (Texts, Than) are unblocked and can proceed in parallel — same audit template, and Sources already has a head start from Spike 5's bibcite profiling.
  • Workstream B (Images interactive UI, Than) remains the largest unstarted piece of Sprint 2 and has no dependency on C.