What is happening
During the #6445 reindex, 88 records dead-lettered out of the 2026-07-03 transformers and stayed there on redrive. All three sets fail deterministically at transform with source-data problems, each hitting a deliberate strictness guard in the transformer. They are three separate problems with three likely owners, described together here because they were found together; split out as needed.
None of this blocks comparison (#6464): 88 known-absent works out of 3.04M, ids listed below.
1. Sierra: MeSH identifiers in Library of Congress fields, introduced by the FOLIO migration
10 bibs fail with Could not determine LoC scheme from id 'D002378' (and similar). Each carries a MeSH identifier (D-prefix) in a MARC field whose indicator claims an LoC scheme. The transformer guard is intentional (MarcHasRecordControlNumber.scala: "Common mistakes include MeSH ids marked as LoC ids... Guard against these by rejecting the identifier"), so these records are miscatalogued.
Two aggravations. All ten were updated on 2026-07-22 with a dl3 folio migration note, so the migration tooling probably introduced the pattern and further batches may add more. And this is live in production, not reindex-only: the production Sierra DLQ currently holds 7 of the same records.
Decisions needed: fix the records (cataloguing or migration tooling), and separately whether the transformer should drop an invalid identifier with a warning rather than rejecting the whole work. The guard is also what keeps miscatalogued ids from leaking into the API as false LCSH links, so softening it is a real trade-off rather than an obvious win.
Sierra bib numbers (10)
2026566211191728490202870262304318130489313287382332023133958733396313
2. METS: born-digital packages without a premis rightsMD element
68 born-digital Archivematica packages fail with Archivematica Mets file must contain a premis-compatible rightsMD element, about 2.6% of born-digital. The strictness was provisional when written; the check in MetsXml.scala carries the comment "I don't yet know if this is strictly true, or whether we can define a default value for any missing ones", and these packages are the case it was unsure about.
Decision needed: either re-ingest the packages with rights metadata, or answer the code's own question by defining a default access condition (presumably closed) for born-digital packages without one. The second is a small transformer change once digital production confirms the default.
METS identifiers (68)
PPCNE/D/3/19PPGRF/A/36/41PPGRF/H/35/36APPJLE/A/30/3PPKIN/A/2/21/2PPKIN/B/2/9/5PPKIN/B/2/9/6PPKIN/H/5PPKZR/E/52aPPKZR/G/8PPMDM/A/1/1aPPMDM/A/1/2aPPMDM/A/1/2bPPMDM/A/2/10PPMDM/A/3/1aPPMDM/A/3/3/1aPPMDM/A/3/4aPPMDM/A/3/5/3PPMDM/A/4/1aPPMDM/A/4/4aPPMDM/A/5aPPMDM/B/20PPMDM/C/1aPPMDM/C/2aPPMYG/A/4/19/4PPROS/K/3/1PPROS/K/3/2SA/CER/A/7SA/REN/B/21/1SA/REN/T/1SA/REN/T/2SA/REN/T/3SA/REN/T/4SA/REN/T/5SABIO/M/8SAHEC/B/20/5SAIPA/A/8/8aSANNS/H/1SAREN/N/3/2SASRH/A/1/2SASRH/A/3/2SASRH/B/40/2SASRH/B/40/3SASRH/B/42SASRH/C/2/3SASRH/D/5SAWFO/J/1SAWFO/J/2/1SAWFO/J/2/2SAWFO/J/2/3SAWFO/J/3SAWFO/J/5SAWFO/J/6SAWFO/J/7SAWFO/J/8b1458606xb20442117b21705938b2170594xb24873342b24875831b31360051b33014772b33508082b33511196b33670912b33671564b33672520
3. TEI: manuscript nodes without an id
10 manuscripts fail with Could not find an id in node! (TeiOps.scala): nodes in the TEI XML missing the id attribute the transformer requires. The TEI files are hand-maintained in the wellcome-collection-tei repository, so this is markup to fix upstream rather than a transformer change; the strictness is guarding real malformation.
TEI manuscript ids (10)
manuscript_16057manuscript_16082manuscript_16094manuscript_16417manuscript_16441manuscript_16460manuscript_16464manuscript_16475manuscript_16514manuscript_16530
Evidence
Diagnosis on #6445 (2026-08-03 comment): every id fails with an identical error across the original run, SQS retries and a redrive, so all three classes are deterministic. The DLQ messages age out after 14 days; the id lists above are the durable record.
What is happening
During the #6445 reindex, 88 records dead-lettered out of the 2026-07-03 transformers and stayed there on redrive. All three sets fail deterministically at transform with source-data problems, each hitting a deliberate strictness guard in the transformer. They are three separate problems with three likely owners, described together here because they were found together; split out as needed.
None of this blocks comparison (#6464): 88 known-absent works out of 3.04M, ids listed below.
1. Sierra: MeSH identifiers in Library of Congress fields, introduced by the FOLIO migration
10 bibs fail with
Could not determine LoC scheme from id 'D002378'(and similar). Each carries a MeSH identifier (D-prefix) in a MARC field whose indicator claims an LoC scheme. The transformer guard is intentional (MarcHasRecordControlNumber.scala: "Common mistakes include MeSH ids marked as LoC ids... Guard against these by rejecting the identifier"), so these records are miscatalogued.Two aggravations. All ten were updated on 2026-07-22 with a
dl3 folio migrationnote, so the migration tooling probably introduced the pattern and further batches may add more. And this is live in production, not reindex-only: the production Sierra DLQ currently holds 7 of the same records.Decisions needed: fix the records (cataloguing or migration tooling), and separately whether the transformer should drop an invalid identifier with a warning rather than rejecting the whole work. The guard is also what keeps miscatalogued ids from leaking into the API as false LCSH links, so softening it is a real trade-off rather than an obvious win.
Sierra bib numbers (10)
20265662111917284902028702623043181304893132873823320231339587333963132. METS: born-digital packages without a premis rightsMD element
68 born-digital Archivematica packages fail with
Archivematica Mets file must contain a premis-compatible rightsMD element, about 2.6% of born-digital. The strictness was provisional when written; the check inMetsXml.scalacarries the comment "I don't yet know if this is strictly true, or whether we can define a default value for any missing ones", and these packages are the case it was unsure about.Decision needed: either re-ingest the packages with rights metadata, or answer the code's own question by defining a default access condition (presumably closed) for born-digital packages without one. The second is a small transformer change once digital production confirms the default.
METS identifiers (68)
PPCNE/D/3/19PPGRF/A/36/41PPGRF/H/35/36APPJLE/A/30/3PPKIN/A/2/21/2PPKIN/B/2/9/5PPKIN/B/2/9/6PPKIN/H/5PPKZR/E/52aPPKZR/G/8PPMDM/A/1/1aPPMDM/A/1/2aPPMDM/A/1/2bPPMDM/A/2/10PPMDM/A/3/1aPPMDM/A/3/3/1aPPMDM/A/3/4aPPMDM/A/3/5/3PPMDM/A/4/1aPPMDM/A/4/4aPPMDM/A/5aPPMDM/B/20PPMDM/C/1aPPMDM/C/2aPPMYG/A/4/19/4PPROS/K/3/1PPROS/K/3/2SA/CER/A/7SA/REN/B/21/1SA/REN/T/1SA/REN/T/2SA/REN/T/3SA/REN/T/4SA/REN/T/5SABIO/M/8SAHEC/B/20/5SAIPA/A/8/8aSANNS/H/1SAREN/N/3/2SASRH/A/1/2SASRH/A/3/2SASRH/B/40/2SASRH/B/40/3SASRH/B/42SASRH/C/2/3SASRH/D/5SAWFO/J/1SAWFO/J/2/1SAWFO/J/2/2SAWFO/J/2/3SAWFO/J/3SAWFO/J/5SAWFO/J/6SAWFO/J/7SAWFO/J/8b1458606xb20442117b21705938b2170594xb24873342b24875831b31360051b33014772b33508082b33511196b33670912b33671564b336725203. TEI: manuscript nodes without an id
10 manuscripts fail with
Could not find an id in node!(TeiOps.scala): nodes in the TEI XML missing the id attribute the transformer requires. The TEI files are hand-maintained in the wellcome-collection-tei repository, so this is markup to fix upstream rather than a transformer change; the strictness is guarding real malformation.TEI manuscript ids (10)
manuscript_16057manuscript_16082manuscript_16094manuscript_16417manuscript_16441manuscript_16460manuscript_16464manuscript_16475manuscript_16514manuscript_16530Evidence
Diagnosis on #6445 (2026-08-03 comment): every id fails with an identical error across the original run, SQS retries and a redrive, so all three classes are deterministic. The DLQ messages age out after 14 days; the id lists above are the durable record.