Skip to content

Session Log: Texts content-model audit (Sprint 2 Workstream C3) — all three C audits closed

Date: 2026-09-01 Participants: Than Grove (in a live group session with Yuji Shinozaki and Xiaoming Wang), Claude Code Outcome: docs/planning/texts-content-model-audit.md written and committed, validated against the existing d7_texts DDEV database. This closes Sprint 2's C1/C2/C3 trio — all three now have real data-profile validation. The og_membership collection-membership finding is now confirmed 3-for-3 and consolidated into one shared deferred note.

1. A head start from an already-closed spike

Unlike AV and Sources, Texts already had substantial groundwork: Spike 4b (CKEditor 5 footnotes) is CLOSED/proven, with a full investigation of shanti_texts' body content and the cross-page footnote citation/definition pattern. That spike's memory ([[project-spike-4b-ckeditor-footnotes]]) supplied node counts, the field_book_content body field, the Tibetan-language count (1,138), and — critically — a d7_texts database already loaded in DDEV, so no dump-hunting or corruption risk this time. This audit deliberately did not re-investigate footnotes mechanics (out of scope, already fully resolved) — only confirmed the module's existence for completeness.

2. Key structural finding: book is D7 core's Book module, not a custom bundle

The one audit of the three where the primary content type isn't site-invented at all. shanti_texts_features_node_info()'s book entry uses base => 'node_content' and Drupal core's own stock Book-module description string verbatim — it's re-asserting the core bundle plus custom fields, not defining a new type. field_book_mlid (a computed, non-stored field deriving from core's menu_links table) confirms the outline hierarchy lives entirely in core's own {book}/{menu_links} tables. This means the outline/hierarchy migration path is a known, core-supported mechanism — a smaller open question than the modeling decisions AV and Sources both face.

3. shanti_texts_splitter clarified — not a migration tool

Read the code directly rather than guessing from the name: it's an editorial convenience that auto-splits one pasted document into a full book-outline tree at node-save time, based on heading levels. Irrelevant to D11 migration logic itself, but its own code independently distrusts field_og_collection_ref's storage enough to explicitly reset it to empty on every generated page ("Without reseting collection for pages, site crashes") — a hint that foreshadowed the data-profile finding below.

4. The og_membership bug — now confirmed 3 for 3

Checked the same collection-membership gap found on both AV and Sources this session: field_data_field_og_collection_ref is empty (0 rows) for Texts too, with real membership (7,419 node memberships, 97.1% of book nodes) living entirely in og_membership. With all three sites checked this sprint showing the identical pattern, this stopped being a per-site quirk worth three separate notes — consolidated into one shared deferred note, og-collection-ref-storage-empty-use-og-membership.md, cross-referencing all three audits and flagging that Images (audited before this pattern was known) should be checked too before assuming migration parity across all four sites.

5. Other real-data findings

  • Three overlapping "language" fields resolved with real numbers: field_dc_lang_code (ISO list) is the clear practical primary at 62.8% filled (matching Spike 4b's known Tibetan count of 1,138 exactly), vs. field_dc_language_original (free text, 3.2%) and field_language_kmap (KMaps taxonomy, 2.3%) — both minority-use.
  • group_content_access (required) is 100% filled — clean, matching AV, unlike Sources' 91.2%.
  • field_split_text has 2,887 rows but all currently value 0 — no books mid-flagged for splitting in this snapshot, consistent with it being a transient authoring toggle.

6. Doc/tracking updates

  • docs/planning/texts-content-model-audit.md — full audit, code + data verified.
  • docs/deferred/og-collection-ref-storage-empty-use-og-membership.md — new, consolidating the cross-site finding.
  • Sprint 2 backlog: C3 row checked off. The sprint's own acceptance criterion for all three audits (real dump, not placeholders) is now checked — all three exist and are data-validated.

Next-session starting point

Workstream C (the three content-model audits) is done. Workstream B (Images interactive UI — OpenSeadragon viewer, sibling carousel, masonry grid) is now the largest unstarted piece of Sprint 2, owned by Than, with no dependency on C. Workstream D (uniform endpoint access docs) is also open and small. Separately: verify whether Images has the same og_membership gap found on all three other sites, per the new deferred note's item #2.