Targeted chemical-identity review and limits

Purpose
This review records source evidence for high-risk structure assignments. It accompanies the runnable consistency check in analysis/check_structures.py. The generated analysis/structure_audit.json identifies the exact dataset hash it checked. It does not certify acceptance by ORD or assert every experimental reagent identity has been independently confirmed.

Primary source locators
ArXiv paper with SI: https://arxiv.org/pdf/2509.11798
  Combined PDF p50 = SI p35, Supplementary Figure35: sulfonamide ligand chart.
  Combined PDF p51 = SI p36, Supplementary Figure36: solvents, bases and Pd precursors.
  Combined PDF p54 = SI p39, Supplementary Figure39: Suzuki ligand chart.
  Combined PDF p55 = SI p40, Supplementary Figure40: solvents, bases and Pd precursors.
Final journal SI: https://ars.els-cdn.com/content/image/1-s2.0-S2451929426001014-mmc1.pdf
  PDF p36: sulfonamide ligand chart; p41 FigureS44: Suzuki ligand chart; p44 section8.5: separate upscale synthesis.
The publisher PDFs and extracted full text are not redistributed here.

Confirmed drawing-based assignments
1. P(tBu)3: arXiv Figure35 explicitly depicts PtBu3.HBF4, while Figure39 depicts free P(tBu)3. Final journal SI p36/p41 retains this chart distinction. The sulfonamide override in compounds.json encodes C12H28BF4P, including the phosphonium proton and tetrafluoroborate anion; the general entry encodes neutral C12H27P. The automated check tests the appropriate assignment in each campaign rather than treating a shared SURF label as sufficient.

2. QPhos: the charts depict P(tBu)2 on one cyclopentadienyl ring and five phenyl groups on the other. The supplied ionic ferrocene representation contains two Cp-anion fragments and Fe2+, formula C48H47FeP, net charge zero. Omission of explicit haptic bonds in this documented representation is not itself evidence of a wrong compound.

3. Pd precursors: the supplied structures retain the expected compositions: Pd(OAc)2 = C4H6O4Pd; [Pd(allyl)Cl]2 = C6H10Cl2Pd2; Pd2(dba)3 = C51H42O3Pd2. The charts' '[Pd(dba)3]' label conflicts with the authors' SURF '[Pd2(dba)3]'; the curated map documents that conflict and follows the SURF identity. Disconnected/ionic organometallic SMILES are representation conventions and should not be interpreted as detailed bonding or oxidation-state measurements. Loading per Pd atom versus per precursor molecule is a separate source-amount question.

4. Racemic and ambiguous trade-name ligands: targeted source/redraw comparisons support the racemic BI-DIME and QUINAP connectivities, P-stereogenic (tBu)PhCPhos, TrixiePhos, para-NMe2 p-DavePhos, para-n-butyl sulfonamide-PPh2 Metamorphos, PhXPhos, and the CF3-rich RuPhos-Hybrid member. No unsupported single-enantiomer descriptor is encoded. A stereo-unspecified SMILES does not by itself establish a measured1:1 racemate; the source names retain the racemic designation where stated.

Unresolved source question: Suzuki phosphine form
The HTE Suzuki chart depicts free P(tBu)3, but the final journal's separate upscale procedure (SI p44 section8.5) explicitly charges PtBu3.HBF4. The scale-up is a different experiment and may deliberately change reagent form. This is not proof that the seven HTE assignments are wrong. The defensible statement is that the HTE assignment follows the chart; definitive bottle identity would require the authors' reagent inventory or equivalent evidence. Do not infer a salt correction solely from the upscale.

Extent of review
No confirmed material identity defect was found in the targeted high-risk source comparisons. The automated checker covers every bundled map SMILES/InChIKey, CAS check digit, actual serialized input identity and identifier, source substrate/product SMILES, reaction SMILES, and available workup-map identities. These tests establish internal and source-file consistency, not experimental truth. CAS check digits and stored PubChem identifiers are not live database verification. The manual review was targeted and must not be described as a fresh exhaustive external adjudication of all81 ligands or all122 named non-substrate entries.

Reproduce after building the dataset
python analysis/check_structures.py
The checker writes analysis/structure_audit.json and exits nonzero on detected mismatches. It does not modify the dataset or source files.
