Skip to content

Session Log: AV15's last real gaps (duration, description translations), a Mandala Home demo, an IIIF data-hygiene finding, and planning for Wednesday

Date: 2026-09-18 Driver: Yuji Shinozaki (with Claude Code); Xiaoming Wang also present in session Outcome: Closed out AV15's two remaining real gaps (duration, multi-language descriptions), shipped as PR #224/#225, both deployed to dev-0. Built a curated 24-link demo on the placeholder home page, catching and correcting a subtle "the node loads but the embedded image doesn't" bug along the way. That bug led to a genuine, well-scoped IIIF/Cantaloupe finding — 10 known test images, not real content, 404 on a legacy S3 key layout that predates the D11 migration entirely. Decided (but did not build) the Mandala Home carousel's architecture. Learned Than's return slipped from Monday to Wednesday and re-planned accordingly.

PRs merged: #224, #225, #226 (docs only)

1. AV15's last two real gaps

Picked up directly from the prior session's "list of all the translations" / "duration is also missing" pushback — both were things I'd earlier written off too quickly, and both turned out to be real, fixable gaps.

  • Description translations. field_pbcore_description is multi-valued, but the Overview block only ever read ->first(), silently discarding every other language. The reference node alone has 6 real items across English/Tibetan/ Chinese. Checked D7's live page directly: it shows the primary description inline and hides the rest behind a .showdesclang "(Show All Languages)" toggle — collapsed by default. Replicated that shape with a native <details>/<summary> (no new JS needed) rather than porting D7's own hidden-class/langname-label markup.
  • Duration. First written off as "read live from the Kaltura entry, out of scope." Wrong: D7 caches it locally, node_kaltura.kaltura_duration, joined via the node's already-migrated Kaltura entry id — a perfectly migratable, static value. Confirmed exactly on a live example (194s → "3 min 14 sec", matching D7's page). New field_kaltura_duration + drush av:backfill-kaltura-duration (mandala_migrations) mirror it into D11; all 11,485 AV nodes backfilled.
  • Along the way: also fixed a real Overview-suppression bug (an early-return guard was hiding the whole block, including the date, on any node with no creator/description/collection — 16 nodes corpus-wide) and re-checked a user report of a node showing nothing under "Video Overview" (turned out to be browser cache from before that fix).

A full-corpus check asked "so are the two duration values ever out of sync?" found real data-quality drift: PBCore's own field_duration is populated on only 17% of AV hosts, and of those, 41% disagree with kaltura_duration — sometimes by a second (rounding), sometimes by minutes. Not a D11 bug (D7's UI never shows field_duration either), but a genuine data-cleanup question for Than/AV cataloging staff, filed as its own doc: av15-pbcore-duration-vs-kaltura-duration.md. Also corrected av15-avinfo-abandoned-fields-review-with-than.md — avduration was originally filed there as a third "likely abandoned" field alongside avuploader/avrating; it's resolved now, those two aren't.

Asked to add sample content to the placeholder home page under the Audio & Video / Images links, showing off real fixes from this sprint. Iterated through several rounds of direct feedback rather than landing on a final list in one pass:

  • Started at 4 AV / 3 Images / 2 groups, then expanded to "about 8 in each section" on request, including a deliberate AV/Images mix in the Collections section (not just generic ones) — picked the flagship AV node's own actual collection/subcollection so the demo page and the node page tell the same story.
  • Told to lean toward Tibetan/Chinese-language content, matching what the AV/ Images corpus actually is, rather than picking arbitrary English-titled examples for readability. Re-picked most of the list around real Tibetan cultural imagery (Lhasa rock carvings, prayer wheels, historic Frederick Williamson Collection photos, Buddhist murals) and Tibetan-region AV content (Bhutan, Amdo, Kham, Rebgong).
  • Every one of the 24 final links was verified live (HTTP 200, and for groups, actually public) before being called done — this is what caught the next finding.

3. Blue Grosbeak: a real image node that loads but doesn't display

User caught that the Blue Grosbeak demo pick (chosen specifically to show off the IIIF/OpenSeadragon deep-zoom viewer) had a missing image, and separately flagged the Hummingbird image as broken too. Investigated properly rather than assuming:

  • Confirmed via the live IIIF endpoint that both 404 — with a leaked Cantaloupe stack trace already documented in a Sprint 1 deferred doc.
  • Checked S3 directly: the JP2 source files genuinely exist (2.46MB, 1.64MB, both from 2020) — not a missing-data problem. Cantaloupe is computing a malformed double-slash key for a flat-layout object.
  • Scoped it properly, not by guessing: a first correlation (missing field_iiif_mms_id) suggested this might be corpus-wide (~38% of all Images nodes) — disproven by testing a wide random sample that all resolved fine. The real, data-confirmed boundary: every real field_iiif_id numeric value ≤100 404s, everything above resolves. Exactly 10 nodes match, all nid 1-10, the very first Images nodes ever migrated.
  • Checked whether this was a wider Banksy-specific problem (user's hunch) — the other 4 Banksy-titled images all resolve fine, so the broken set stays exactly those 10.
  • Confirmed against live D7 production directly: the same broken iiif.lib.virginia.edu URL is embedded verbatim in D7's own page markup. Not a D11 regression — this predates the migration entirely.
  • User confirmed directly: all 10 are known test/dev images from an early ingest batch (John Alexander), not real archival content. Rewrote iiif-cantaloupe-404-information-disclosure.md accordingly — reframed from "10 real broken images, ask DevOps to fix the key delegate" to "a data-hygiene question first" (delete the test nodes, which makes the infra bug moot for Mandala entirely), infra fix a distant second. Dropped Blue Grosbeak from the demo list; the other 7 Images picks already had verified real IIIF data, so no replacement was needed.

4. A process-management mistake, caught and fixed

Launching the dev-0 backfill, a first attempt used a shell & backgrounding trick, judged (wrongly) not to have survived; a second, properly-tracked attempt was launched on top of it. Both actually reached dev-0 and ran concurrently — two writers contending for locks on the same table, which explained a ~20x slowdown the user noticed and asked about directly. Diagnosed via ps aux on the container (found two real drush av:backfill-kaltura-duration processes, both started at the same time), killed both, and relaunched a single run properly detached this time (docker exec -d, output to a log file inside the container) so it survives independent of this session and can be checked via a simple row-count query rather than continuous monitoring.

5. Shipped: PR #224, #225, #226

  • #224 — the duration field/backfill, description translations, Overview- suppression fix, and the first pass of the home-page demo. Merged, deployed clean (config:status gate passed).
  • #225 — dropped Blue Grosbeak from the demo + the corrected IIIF deferred doc. Merged, deployed clean.
  • #226 — the Mandala Home architecture decision (see below), docs-only, no deploy triggered (confirmed against the trigger_paths fix from an earlier session — only drupal/**/package/**/pipeline/** changes deploy).

Both real deploys were watched through to Deploy by pipelineExecutionId, not latestExecution status alone (the same known trap as before — Deploy's status lags behind which execution it belongs to).

6. Planning for Wednesday, not Monday

Mid-session, learned Than's return slipped from Monday 2026-09-22 to Wednesday 2026-09-24. Re-derived what's actually available in the gap from the real project docs rather than guessing:

  • Genuinely still blocked on Than: AV7's two remaining access realms (group_access_uva_member, mb_collection_admin — nobody's read the D7 mechanism yet), AV11→AV12 (Kaltura upload track), Texts (his own track), and six deferred docs now queued for his review.
  • Worth reconsidering given two extra days: Texts/Sources were paused 2026-09-04 as a team capacity decision, not because Sources specifically needs Than — that's Xiaoming's own track. Not decided whether to resume it, but noted as an open question in docs/roadmap.md rather than left unexamined.
  • A real gap in the first pass at this: initially didn't mention AV transcripts (Sprint 4) at all. Corrected once asked — Sprint 4 is blocked on Spike 11, still ○ Pending, individually owned by Than. Real prior investigation already exists in that spike doc (D7's transcript pipeline — Toolbox/SRT/XML → Java Saxon XSLT → multi-tier TCUs — already reverse-engineered from reading the real module code), so it's not starting from zero, but it's a substantial spike with a working-prototype pass criterion, not a quick task, and one of its central scope questions may need David Germano, not just Than.

7. Mandala Home: decided, not built

Asked to start on Mandala Home's real content system (currently a placeholder linking to /images//av). Before writing code, surfaced the real open architecture decision already flagged in mandala-home-customizable-content-system.md (Block Plugin vs. Layout Builder vs. a real content entity for the hero carousel) and asked directly rather than assuming. Decided: a mandala_home_slide Paragraph type + a mandala_home_carousel custom block_content type referencing it — same structured-Paragraphs pattern already used everywhere else in this project, not a new mechanism. The two static feature panels use core's existing Basic block type. Mid-build (had started scaffolding the Entity API script), told to hold off — planning only, implementation starts Monday. Recorded the decision in the deferred doc and shipped that as its own docs-only PR (#226) rather than leaving it only in this log.

What's next

Monday: the regression sweep (deferred from this session so the demo/duration work could ship first), then start on Mandala Home per the decision above. Whether Sources resumes before Wednesday is an open team question. AV transcripts (Spike 11) is a real, substantial candidate for Monday/Tuesday if someone wants to run it directly rather than wait for Than, with the caveat above about David Germano. Wednesday: everything queued specifically for Than — AV7's two realms, AV11→AV12, Texts, and six deferred docs.