The annotation checkpoint evaluates 70 imported source contexts, retaining their order, identifiers, evidence, source references, qualifiers and supporting entities. Seventy-one raw QuickGO records become 70 normal rows because two exosome records collapse when the normal TSV omits their differing raw annotation ids and one cell-type extension. Both raw records remain preserved in the acquisition evidence; no annotation or product is silently deleted.
The review has 53 ACCEPT, eight MODIFY, three KEEP_AS_NON_CORE, one REMOVE and five UNDECIDED decisions. Broad membrane rows are refined to plasma membrane; generic hydrolase is refined to apyrase; coagulation/platelet-aggregation rows are refined to negative regulation of platelet aggregation on the independent endothelial ADP-hydrolysis experiment. NDP/NTP class activities are retained because the human substrate panel establishes breadth. A single-substrate experiment does not require narrowing the whole-gene class.
Supported RanBPM binding is retained as non-core without assigning CD39 an unsupported adapter function. Caveolar enrichment is also non-core. The remaining uncertainties concern cell adhesion, CDP-specific activity, two narrow UDP donor/Rhea chains, and the unrecovered ENTPD1 entry in the exosome supplementary table. The separate PAINT UDP assertion is accepted on actual ancestral placement; no human UDP assay is claimed.
The original-evidence guide provides public URLs and precise read limits. Independent originals resolve source-only access gaps where they establish the same biological claim. An ACCEPT or MODIFY does not mean every cited original was fully available. Six explicit reviewer full-text-unavailable flags describe actual access limits; recovered originals are not mislabeled from the normal fetcher's cache flags.
Two unreported source defects are recorded: the incorrect DOI in UniProt's PMID8626624 reference block, and a request to check mouse basement-membrane localization. The latter follows actual Methods/Results/captions and an independent peer review. It does not imply wrong orthology, gene-name confusion, a universal absence of matrix association or a new NTPDase2 annotation. Original source fields and caches remain unchanged.
The six natural products are preserved separately. Native placental protein preparations, 517/306-residue placental cDNAs, and synthetic human/rat deletion constructs are distinct. Sequence reconstruction matches each independent FASTA; loss or retention of a segment is not alone a functional assay.
The normal source set contains sixteen accepted publication caches plus three genuine donor-paper additions. The genuine Falcon/perplexity-lite research invocation failed during bootstrap; no provider report was fabricated. Curated interpretation uses the inspected originals and database provenance instead. No new supporting-text quotations are added.
The completed review includes one independently synthesized apyrase core, connecting successive ATP/ADP hydrolysis and inorganic phosphate release to cell-surface nucleotide catabolism and a contribution to inhibition of platelet aggregation. AMP-to-adenosine conversion belongs to CD73. PMID:8529670, PMID:8996251 and PMID:10636909 provide the directly inspected core evidence; PMID:8955160 remains abstract-only. The six products are not presumed to share every catalytic or localization property. The final normal validation and preservation receipts are kept separately; DRAFT status reflects remaining advisories and is not a statement that any imported row is unreviewed. Publication and campaign credit require separate verification.
The reproducible product analysis includes exact input snapshots, a standard-library reconstruction script and independent human/rat FASTA comparisons. Its unchanged-residue projections do not predict activity or localization.
The UDP IBA/IEA/ISS split is intentional and now has explicit consistency exceptions on the two unresolved source routes. The independently verified PAINT judgment supports inherited UDP activity; the narrower rat-paper/Rhea routes remain incompletely adjudicated. These are source-access limits, not contradictory assertions that UDP hydrolysis is absent. The B-cell adhesion, human CDP/UDP kinetics and GTPase-terminology questions are now explicit, with two focused experimental designs. They do not claim an exhaustive search established a universal knowledge gap.
The description includes the contribution of ATP/ADP removal and AMP supply to immune regulation. The inspected mouse T-cell study, PMID:17502665 Results sections3–5/Figures3–5 and Methods12–15, separates CD39 from downstream CD73 and adenosine-receptor actions; its unpublished human observations are not treated as replication. The two placental references now carry bounded findings explaining their product-context relevance.
Supported RanBPM association remains non-core under the campaign curation instructions. The primary assay makes RanBPM the regulator of CD39 and does not establish CD39 adapter or enzyme-regulator activity. The same campaign instructions retain reference-only evidence with precise locators and prohibit new publication quotations; they intentionally differ from the general binding and quotation recommendations. All 70 source objects, actions, six products, normal caches and core functions are unchanged.
The product replay now computes its exact-match report fields from sequence comparisons. Assertions remain in place, and the human/rat result files are unchanged.