The starting review matched the GitHub main blob 5c662478ff2a9c602f79d2e6b31d7b32f3a79620 at main 6455f59face473d84f2976a81c54aa47df26f398. An open-PR search found no ACAD9 conflict. The starting file contained 40 source assertions and one author-proposed NEW annotation, no NOT assertions, and two core functions. All 40 source tuples are preserved, including their original term, evidence code, reference and any seeded flags. The prior NEW proposal is withdrawn as explained below.
This is a substantive re-review under the review, annotation-reviewer and core-function-synthesizer skills. The assigned agent serves as the annotation reviewer. The existing genuine Falcon report (2026-02-11; 318.93 seconds) was read and left unchanged. The campaign coordinator explicitly assigned an audit using available caches while new-source seeding is unavailable; no new provider report was manufactured. The generated OpenScientist hypothesis and its artifacts also remain unchanged.
All original PMID abstracts/caches and the five Reactome summaries were inspected. Full cached primary texts were available for PMID:33320993, PMID:33753518, PMID:34646991, PMID:34800366 and PMID:38086790. PMID:25721401 was already present elsewhere in the shared publication cache with full text and has been added to the review's references without changing that cache.
Supported fetch-pmid --force --output-dir /tmp/acad9-fetched attempts for PMID:21237683, PMID:32320651 and PMID:16020546, plus a fetch of PMID:25721401 to that temporary directory, all terminated with DNS error nodename nor servname provided, or not known. No generated cache content was replaced. Additional primary access was possible through the public web reader:
PMID:16020546, PMID:20816094 and PMID:24158852 remain abstract-only locally. The 2005 publisher page did not expose the full substrate/kinetic tables. The curated UniProt record supplies independent provenance for medium-chain reactions, but was not relabeled as newly verified primary assay evidence. PMID:34800366's ACAD9-specific supplementary record was not re-extracted: the mitochondrial annotation is retained with explicit curator deference and independent fractionation support.
Long-chain enzyme specificity. The GO:0004466 definition uses 13-22 carbons; GO:0017099 uses more than 22. The two 2021 source assays use C16:0-CoA: ETF fluorescence reduction in PMID:33320993 and ferricenium reduction in PMID:34646991. Their MODIFY to GO:0004466 therefore remains justified by the tested substrate. It does not establish universal inability to oxidize longer chains. The earlier C20+ cutoff and claims that preferred substrates exclude other activity were removed. The OpenScientist cavity-residue argument is an unvalidated computational hypothesis, not evidence for a negative biochemical claim. Its reference assessment is now UNVERIFIED for that mechanistic conclusion.
The 83 min-1 and 18%-of-VLCAD passages in PMID:34646991 concern a particular purified-protein assay and FAD state. They are not a universal activity ratio, physiological importance ranking, or GO chain-length test. Precise 2005 kinetic constants were omitted because the original table was not recovered. Its accessible abstract directly supports the tested unsaturated C16:1, C18:1, C18:2 and C22:6 substrates.
Two biological roles. PMID:25721401 directly tests human ACAD9 knockout and rescue, not merely disease association. Its HEK293 results support a cellular long-chain FAO contribution alongside complex I assembly. The variant/clinical correlation is kept distinct from direct causal quantification in affected human tissues. The earlier blanket claim that disease mutations do not affect FAO is withdrawn; PMID:20816094's negative observation is scoped to its models. The long-chain metabolic-process row is now ACCEPT, consistent with the enzymatic core. PMID:24158852 supports partial-to-complete assembly rescue by catalytically inactive ACAD9, rather than an unqualified statement that every assembly property is unaffected.
Medium-chain MF and process annotations remain NON_CORE, with deference to the experimental curation and matching UniProt catalytic assignments for C9-C11 substrates. The uninspected full substrate panel and lack of a demonstrated major physiological medium-chain role are stated explicitly. No quantitative superiority of MCAD is claimed.
Assembly mechanism. PMID:32320651 supports ACAD9-centered MCIA stability and assembly-intermediate association. In PMID:33320993 the resolved bridge between ACAD9 and NDUFAF1 is ECSIT, and no direct ACAD9-NDUFAF1 interaction was observed. The GO:0030674 definition requires bringing macromolecules together and distinguishes scaffold activity. Those data do not settle ACAD9's own precise adaptor mechanism. Its source IDA is preserved with UNDECIDED; this is uncertainty, not a declaration that the curator is wrong. The assembly core retains the established process, location and MCIA membership while leaving a precise MF unassigned. No scaffold function is substituted without evidence.
All three generic protein-binding annotations are REMOVE under the binding policy. The evidence for interaction or co-association is retained; removal means the generic term is uninformative, not that the interaction is false. The PMID:33753518 partner description also no longer misidentifies Q9BQ95 as TMEM126B.
Neuronal location. The 2011 paper explicitly describes dendritic ACAD9 staining; that row is NON_CORE. Mitochondrial residence does not invalidate a neuronal-compartment observation. Cellular-nucleus localization is UNDECIDED: the accessible sections distinguish ACAD11 nuclear staining and an anatomical dentate nucleus from the GO cellular nucleus. The old unsupported overexpression/cross-reactivity explanation is removed. No misattribution is asserted from incomplete access.
Propagation and pathway context. Ten IBA/IEA reviews now trace proximate sources. PAINT is recorded at PTN000744030; self-inclusion among descendant evidence is not circularity. InterPro comments use the domain descriptions in the cached UniProt record, without converting domain membership into substrate-specific proof. ARBA rule internals remain UNRESOLVED while independent activity/localization evidence can support acceptance. Rhea and EC sources are traced to the curated catalytic record. The five Reactome location rows retain assembly context and do not imply mature complex I catalytic membership.
Prior NEW proposal. ACAD9 does directly perform a beta-oxidation step; participation is not the problem. Nevertheless, the prior author-created NEW GO:0006635 duplicates coverage below existing GO:0006631 in the verified ontology hierarchy. It is withdrawn under the current ancestor/descendant redundancy rule, not converted into a source assertion. No ACAD9/Q9H845 entry was found in gocams/index.tsv. No new process annotation is proposed. The enzyme core retains precise GO:0006635 because direct catalytic participation is established; the validator advisory about its absence from existing_annotations is intentional and does not warrant creating a redundant NEW row.
Additional primary lead. The cached UniProt record cites PMID:40255280, Fabbian et al. 2025, for in-vitro NMR/MS evidence of beta-gamma dehydrogenation products alongside the expected alpha-beta product. This is a follow-up question about product specificity and physiological relevance, not grounds for a new physiological process annotation in this cache-based audit. It is not used to replace the established long-chain dehydrogenase core.
Source-tuple equality was checked against the saved baseline: all 40 source assertions are identical and in the same order. Only the prior NEW row was removed. Targeted schema, ontology, reference and best-practice checks, history validation and rendering are recorded in the matching append-only history after completion. No UniProt, GOA, provider, hypothesis, publication, Reactome or GO-CAM cache was manually edited.
Independent read-only review by annotation_aars2 covered all 40 source assertions and both cores and found no biological blockers. The reviewer independently inspected the 2020 PDF, 2021 bridging evidence, 2015 FAO results, relevant GO definitions and the 2011 dendrite section. Its suggestion to remove an unverified disease-variant domain-clustering clause was adopted without changing the assembly ACCEPT decision.
Responding to the external review on the published dcf2389b45 head, the existing GO:0006631 ARBA row is now MODIFY to GO:0006635 fatty acid beta-oxidation. Withdrawing a redundant NEW row remains appropriate, but it does not prevent refining the existing broad assertion. The replacement is justified by ACAD9's own dehydrogenation chemistry and the human knockout/rescue FAO result, rather than solely by a validator advisory. This resolves the earlier core-to-annotation specificity gap while preserving all 40 source assertions and adding no NEW row. The ARBA rule predicates remain uninspected, explicitly separated from the primary evidence for the refinement.
The PMID:21237683 reference now explicitly flags unavailable local full text. Dendritic staining is attributed to the externally read PMC3073726 section 3.8, with an access-scope explanation in each neuronal row and the reference review. The cached granular-layer sentence was removed from both rows' supporting_text because it does not establish either dendrites or a chromosome-containing nucleus. The dendrite decision remains NON_CORE and the cellular-nucleus decision remains UNDECIDED. No publication cache was manually reconstructed or altered.
The requested comparator check inspected the unchanged human ACADVL curated record and its GO:0017099 IDA assignment to PMID:9461620. That primary abstract describes palmitoyl-CoA assays for active-site/FAD mutants. This supports the reviewer's concern that existing annotation practice may sometimes reflect the historically named VLCAD class, whereas the live GO definition specifies substrates longer than C22. The two ACAD9 MODIFYs remain anchored to their demonstrated C16 reactions; they do not assert inability to use longer substrates or claim the ACADVL annotation is wrong. An expert question now records the class-versus-chain-length convention explicitly. The comparator article is already cached, and its reference entry makes clear that it does not supply ACAD9 evidence. No comparator gene file was edited.
The assembly core now records its unresolved precise molecular-function assignment under knowledge_gaps. The boundary is the established assembly dependence, ACAD9-centered stability hierarchy and ECSIT-mediated connection/deflavination; the examined studies do not by themselves establish a direct ACAD9 adaptor/scaffold action. This is scoped to the evidence examined, without claiming that all molecular details of MCIA are unknown. The existing Falcon file is declared in references as a genuine secondary research artifact, and the description's closing rebuttal clause was removed to leave a standalone biological summary.
Final action counts are 28 ACCEPT, four KEEP_AS_NON_CORE, three MODIFY, three REMOVE and two UNDECIDED. All 40 seeded assertions and all evidence/source files remain unchanged; the change adds no assertion and does not amend the earlier history record. Validation and render results are recorded in the new follow-up history and publication manifest.