Skip to content

AV5 / AV14: Dispositions for the two anomalous AV node populations

Decision date: 2026-09-08 Decided by: Yuji Shinozaki (Sprint 3 lead, per ADR 018) Backlog items: Sprint 3 AV5 and AV14 Status: Both decided — AV4's source queries can now be written Relates to: ADR 008 (migrate, not improve), AV Content-Model Audit (open question #8; the required-field gap in its data profile), Spike 7 (create_node_mediabase() root cause), AV2 scope note (same investigation)

Sprint 3's acceptance criteria require that these nodes be "triaged with a documented disposition (not silently dropped or silently included)." This note is that document. Both nid lists are recorded in full below so the exclusion and the cleanup list are auditable after the fact.

All figures are from mandala-prod-av-db_2026-09-01.sql.gz (gzip -t verified before use, per Spike 7's dump-hygiene rule), read by streaming the dump. Row counts for field_transcript (5,380), field_subject (16,999) and field_workflow (11,465) reproduce the audit's independently-derived figures exactly, which cross-validates the extraction.

AV5 — the 68 MISSING_TYPE nodes: exclude

Decision

Exclude all 68 from migration. Do not create a D11 bundle for them, do not repair them to audio/video. Record the nid list (below) in the migration source query's exclusion so the omission is explicit rather than a side effect of filtering by bundle name.

Why the alternative isn't available

The audit framed this as "exclude, or repair to a real bundle by checking the underlying Kaltura entry's actual type." Repair is not on the table, because there is nothing to repair from:

  • They carry no field_video, no field_audio, and no thumbnail. The only field instance defined on the MISSING_TYPE bundle is og_group_ref. When create_node_mediabase() wrote the invalid bundle string, the Field API had no matching instance to attach the media reference to, so the Kaltura entry ID was silently dropped at save time. There is no entry ID left on these nodes to look up.
  • node_type has no MISSING_TYPE row. It was never a content type — only an invalid string in node.type — so there is no D11 bundle to migrate them into without inventing one.

Why exclusion is safe

Property Value
Count 68, all published
Created all in 2014, all by uid 1
nid ranges 2799–2833 and 2901–2934 — two contiguous runs, consistent with two batch-import passes
Titles all 68 are plain .jpg filenames (100% .jpg; none deviates from an image-filename pattern)
field_pbcore_title 0
field_workflow 0
field_subject / field_kmap_terms 0 / 0
field_transcript 0
field_thumbnail_image 0
Collection membership all 68 in exactly one collection — "Admin: On Kaltura Not in Mediabase" (nid 2503)

That collection is the decisive fact. It is a staff triage bucket whose name states its own purpose, created 2014-11-25 by uid 1. These are Kaltura IMAGE entries pulled through the batch importer and parked there — confirming Spike 7's code-level root cause with data rather than inference.

Excluding them does not strand the collection. Node 2503 holds 353 members: 193 video, 92 audio, and the 68 MISSING_TYPE. The 285 real AV nodes in it migrate normally, so the collection survives with its genuine content intact.

There is no metadata to lose, no media to lose, and no user-facing content: 68 nodes whose entire substance is an image filename.

The 68 excluded nids

2799 2800 2801 2802 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816
2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833
2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917
2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934

AV14 — the 18 AV nodes with no Kaltura entry: migrate as-is, hand staff the list

Decision

Migrate all 18 exactly as D7 has them — published, in their existing collections — and deliver the list below to AV staff as a pre-cutover cleanup item.

This is ADR 008's faithful floor. These nodes are already published and already render metadata with no player on the live D7 site; migrating them as-is reproduces current behaviour rather than making anything worse. Unpublishing them would be us making a content decision on staff's behalf and would change user-facing state relative to D7. Recording the list is what keeps this from being a silent inclusion.

Why not exclude

These are not empty shells like the MISSING_TYPE nodes — they are real records that lost, or never got, their media:

  • All 18 have a field_pbcore_title.
  • 17 of 18 have field_workflow cataloguing/QA data — staff did work on these.
  • 3 carry field_subject KMaps tagging; 1 has an attached transcript.
  • They are spread across 15 different collections, all published.

Only 7 of the 18 sit in obvious scratch collections (test ×2, Docs Test ×2, Create a Collection Test 1, Testing New Collection, test course). The other 11 are in real, substantive collections — Amdo Collection, Tibetan and Himalayan Library, Repgong Folk Songs Project, Hurricane Harvey, White House Evening of Poetry, Music, and the Spoken Word, and four course collections. A blanket exclusion would silently drop catalogued records from real collections on the strength of a missing media reference alone, and splitting the population by collection would require us to judge which collections are "real" — a call AV staff are better placed to make than we are.

The cleanup list

Node titles are deliberately omitted: several are individual presenters' names, and this repository is public. Two course-collection titles have likewise had an instructor surname trimmed (6371 and 15536), so those two strings will not match D7 verbatim — match on the collection nid, not the title. The nid is what makes the list actionable.

D7 nid Bundle Created Collection (nid) Workflow Subject Transcript
1894 video 2012-09-26 Amdo Collection (1760) ✓ — —
2096 video 2014-02-26 Tibetan and Himalayan Library (3) ✓ — —
3895 video 2015-11-12 Create a Collection Test 1 (3888) ✓ — —
3915 video 2015-12-02 Testing New Collection (3914) ✓ — —
4068 video 2016-01-07 Repgong Folk Songs Project (3939) ✓ ✓ —
5581 video 2016-09-04 Italian 2010 (5576) ✓ — —
6196 video 2016-09-29 test (4884) ✓ ✓ —
6211 video 2016-10-13 test (4884) ✓ ✓ —
6566 video 2016-11-10 Section 20 (12:30pm) (6501) ✓ — —
6596 video 2016-11-14 Section 18 (11am) (6386) ✓ — —
6751 video 2016-11-16 STS4500 Fall2016 Prospectus Presentations (6371) ✓ — —
7826 video 2016-11-29 STS4500 Fall2016 Prospectus Presentations (6371) ✓ — —
13766 video 2017-04-24 STS4600 Spring 2017 (15536) ✓ — —
13851 video 2017-07-07 White House Evening of Poetry, Music, and the Spoken Word (13856) ✓ — —
13936 video 2017-07-26 test course (13931) ✓ — —
14561 audio 2017-09-07 Hurricane Harvey (14551) ✓ — —
14636 video 2017-09-08 Docs Test (14626) — — ✓
16246 video 2017-12-04 Docs Test (14626) ✓ — —

Nothing in this population is newer than 2017-12, which suggests whatever caused the media reference to go missing stopped happening some years ago rather than being an ongoing fault.

Note for AV4

field_video/field_audio are required at the D7 form level, and these 18 violate that. The D11 migration will create them the same way (the migrate API bypasses form validation), which means the resulting D11 nodes will also be uneditable in the node form until someone supplies a media reference or the field is made optional. That is a real consequence of the "migrate as-is" choice, and it is the practical reason the cleanup list needs to reach AV staff rather than sitting in a doc.

Observation, not a claim: the Solr AV document count

ADR 016 cites 11,537 production kmassets docs at asset_type: audio-video, against 11,583 audio + video nodes — a gap of 46. Neither population in this note accounts for it: the 68 MISSING_TYPE nodes are a different bundle (and would not carry that facet), and the 18 media-less nodes are the wrong size. The two figures may also simply have been taken on different dates.

This is recorded as a loose thread, not as a finding — it was not investigated here and neither disposition above depends on it. If AV8 (Solr/kmassets sync wiring) wants a reconciled count, this is the discrepancy to start from.