While auditing the existing review, just validate / ai-gene-review validate --terms
flagged an ERROR: the references[].findings entry for PMID:36596815 ("The mitochondrial
Hsp70 controls the assembly of the F(1)F(O)-ATP synthase.") carried a paraphrased
supporting_text ("Mitochondrial Hsp70 cooperates with the assembly factors Atp11 and
Atp12 to form the F1 domain.") that is not a verbatim substring of the cached publication.
The cached record's actual text is: "...it cooperates with the assembly \nfactors Atp11
and Atp12 to form the F1 domain of the ATP synthase." PMID:36596815 — this
exact quote (bridging the line-wrap with ...) was already used correctly
elsewhere in this same review's existing_annotations entries for GO:0033615. Replaced
the paraphrase with this verbatim quote so the references[].findings entry now matches
the same evidence already cited correctly in the annotation review.
Note on severity: PMID_36596815.md has full_text_available: true, and the validator
escalates a non-verbatim finding quote to ERROR only when the cached record is not
abstract-only. That is why this one reference errored while others merely warned — the
severity difference reflects cache completeness, not whether the quote was defensible.
Follow-up to PR review feedback. The four remaining WARNINGs (PMIDs 1532796, 2142305,
10681564, 12829692) were the same defect as the one fixed above — paraphrases in
references[].findings[].supporting_text — downgraded to WARNING only because those four
records are full_text_available: false. The earlier note claiming the source text was
"not present in the cached abstract text" and that this was "expected per project
convention" was wrong on both counts: I re-read all four cached abstracts and the verbatim
text is present in every one, and in each case the correct verbatim quote was already in
use elsewhere in this same review file. There is no convention permitting paraphrase here.
Replaced each with the verbatim string already validated in-file:
F1 vs the abstract's F(1))No other changes: the rest of the review (annotation actions, core_functions, description)
is well-supported by the cited evidence and was left untouched. The review now validates
with zero warnings.
supported_by quote, GO:0051082 action, file: provenance)Addressing the second round of PR review feedback.
1. Non-verbatim supported_by quote (blocking). The earlier sweep covered
references[].findings[] but missed existing_annotations[].review.supported_by. The quote
backing the REMOVE of GO:0005515 read "we generated a high-quality proteome-wide
inter-interactome network map", which occurs nowhere in publications/PMID_27107014.md
(full_text_available: true; grep -c for both "proteome-wide" and "high-quality" returns 0).
Replaced with the verbatim text PMID:27107014. The substance of the
REMOVE is unchanged and still sound: all five GOA rows are IPI/PMID:27107014 differing only in
WITH/FROM, and every partner is a human accession.
2. GO:0051082 now KEEP_AS_NON_CORE, not MODIFY. Both experimental unfolded-protein-binding
annotations were being replaced by GO:0140777. That discards a well-grounded IDA and diverges
from this repo's recorded gene-specific determination in projects/UNFOLDED_PROTEIN_BINDING.md
(ATP11 listed as NON_CORE for UPB). The deciding evidence is that Atp11's chaperone activity is
demonstrated on a non-client substrate: PMID:12829692. Suppressing aggregation of reduced insulin is the canonical unfolded-protein-binding
assay and is not client-specific complex stabilization, so GO:0140777 does not subsume it. Both
GO:0051082 entries were set to KEEP_AS_NON_CORE (setting only one triggers a validator warning
about inconsistent actions for the same term). GO:0140777 remains the core_functions MF.
3. file: reference quotes made verbatim. Five supporting_text values on file: references
were paraphrases. These are never flagged (the file: prefix is in skip_prefixes), but the
schema asks for exact substrings, so they were replaced with real substrings of their sources —
ATP11-deep-research-falcon.md (three) and PTHR13126-metadata.yaml (two, now quoting
accession: PTHR13126...name: CHAPERONE ATP11 rather than a sentence the file does not contain).
4. Curation commentary removed from description. The closing sentence compared annotation
informativeness, which CLAUDE.md reserves for review.reason/notes. Replaced with the biological
fact that motivates the GO:0051082 decision (in vitro holdase activity on reduced insulin).
just validate yeast ATP11 → ✓ Valid, zero warnings.
Not addressed here (out of scope for this PR, flagged by the reviewer): UNFOLDED_PROTEIN_BINDING.md:538
describes ATP11 as the "Atp12p assembly factor", but Atp11 handles F1 beta/Atp2 and Atp12 handles
F1 alpha/Atp1. That page needs a separate fix.