AV4: 84 hosts have genuine content differences between D7's und/en layers — never reviewed
Area: migration / AV / content model / data fidelity
Raised during: Session 2026-09-09 (building AV4), never revisited since
Jira: (add when available)
Priority: Low — out of scope for D11 implementation (the migration
already picks en as authoritative, per design, and that choice is sound);
this is a data-quality question about the D7 source, not a D11 defect
What happened
While confirming AV4's und/en storage-layer handling (see the project's
language-layer background: und is an artifact of an earlier conversion to
language-specific storage, en is the authoritative later layer), the
investigation specifically checked whether und ever duplicates en
content for the same item, as opposed to just linking it at a different
delta. It confirmed the common case is clean: every redundancy found was the
same field_collection item linked twice, not conflicting content.
But 84 hosts were found with genuine content differences between the two
layers — cases where und and en don't just disagree on position, they
disagree on the actual field values. These were flagged for AV staff review
at the time and never followed up on; no list of which 84 hosts, or what the
differences actually are, was preserved anywhere durable (only the count,
in a session-log narrative).
Why this matters
The migration's en-preferred rule means these 84 hosts' D11 content
reflects whichever value happened to be in the en layer — which is the
right default per the established authority ordering, but for these
specific 84 hosts nobody has confirmed that en's value is actually the
correct one rather than, say, a stale or incomplete edit. This is squarely
a D7 content-quality question (was en written correctly for these 84
items?), not something D11's migration logic can resolve on its own.
Recommendation
Re-derive the list of 84 affected hosts (the original query isn't preserved
in the repo, only the resulting count) and hand it to AV staff / Than for a
content review: for each, confirm whether the migrated (en) value is the
one that should have won, or whether the D7 source itself has a data-entry
problem worth correcting before or after cutover.
Related
docs/planning/av-node-migration-notes.md §8;
av4-paragraph-ordering-ties.md (the
sibling ordering-only question from the same investigation);
av4-field-pbcore-language-iso639-conversion.md,
av2-field-tags-old-label-inconsistency.md.