ORD dataset for Grosjean et al., Angew. Chem. Int. Ed. 2025, 64, e202424373

2258 reactions, one ORD Reaction per reaction well, covering all nine iteration/series arrays of the B-SPA campaign. No dataset or reaction IDs; DOI 10.1002/anie.202424373 and the publication URL in every reaction's provenance.

Coverage

Each row is a telescoped sequence on a crude mixture, so it is modelled as one Reaction with conditions_are_dynamic, the full ordered workups chain, and the authors' procedure prose in notes.procedure_details — not as invented per-step reactions the data cannot support.

Conditions are not in the source table. They were joined in per iteration from §7 of the article's Supporting Information (open access, Europe PMC PMC12001203) and the reaction_conditions*.csv files in the deposited Opentrons archive: equivalents, stock concentrations, dispensed volumes, scales, standing times, and the real workup chain. All chemistry is ambient temperature/pressure under air, sealed and unstirred.

Outcomes are qualitative — a CUSTOM measurement carrying detected (1019) / not detected (1024) / trace (27) / inconclusive (65) / not measured (123), with the verbatim source string in details; IDENTITY is added only where the product was seen. to fill, QC not found and N/D are encoded as not measured, never as failures. 1625 rows also carry the authors' numeric yield (824 non-zero) — more than the card suggests.

Row selection

2262 − 4 reactant-less control wells = 2258. The paper's headline 1876 is 2262 − 200 (it 4.1) − 182 (it 5.x) − 4 controls, confirmed independently by Table 1; iterations 4.1 and 5.x are kept because the deposited data reports them, and each reaction carries included_in_paper_headline_1876, asserted to be exactly 1876. No deduplication: 100 of the 113 duplicate reaction SMILES pair an iteration-4.1 well with its iteration-4.2 retry, which is the comparison the paper draws.

Four source defects found, repaired and documented

1. All 512 iteration-3.0 rows have reactant name/SMILES columns shifted one left. Curating as-read would write the string Imid-04 into a SMILES field 512 times and drop 512 carboxylic acids. The repair is verified against Compound uniqueID in 512/512 rows with zero mismatches. Iteration 3.0 is a three-component sequence, not two-component as the card's "Reactant 3 SMILES in 1746 rows" implies.
2. The fourth reactant of iterations 4.1/4.2 (the sulfonamide-forming amine) has no column in the source table; recovered from the compound IDs plus the Opentrons reagent tables, and all 360 verified by substructure against the products.
3. The ferrocene product SMILES lost their iron and one Cp ring (18 rows). Source value kept; a charge-balanced correction supplied as a labelled CUSTOM identifier.
4. Iteration 3.0's core reagent is drawn Boc-protected, contradicting the SI procedure and the products. The only place a source structure was replaced; the literal CSV value is preserved as source_csv_smiles.

Also recorded, not smoothed over: the paper's success totals disagree with each other and with the table (abstract 1108/59%, text 1077/57%, table 1019 success = 54%).

Checks

• 6954/6954 reactant structures pass a scaffold-survival test (MCS vs product, with per-iteration leeway for the expected leaving groups), zero flagged rows; all product SMILES parse; no reactant equals its product.
• ORD validator (ord-schema 0.8.3): 712 warnings, 4516 errors — all 4516 are two messages per reaction from one deliberate choice, leaving record_created unset so that no personal email is baked into 2258 public records. validation.txt includes a second pass of the same build with a curator identity supplied: 0 errors, proving nothing else fails validation. The build script takes CURATOR_EMAIL/CURATOR_NAME/CURATOR_ORCID from the environment, so the submitter fills this in with no code change. The 712 warnings are all the advisory "workup missing inputs" on CONCENTRATION/EXTRACTION steps, which correctly add no species.

Known gap

Iterations 5.1/5.2 (182 reactions) have no published procedure — no SI section, no Table 1/2 entry, no Opentrons protocol directory. Their catalyst, oxidant, base, solvent, temperature and time are left unset with conditions.details saying so, and amounts are explicit UnmeasuredAmounts, rather than guessed. Structures and outcomes are encoded.

One caveat on process: the revision's pinned ord-curation Skill could not be read — every resources/read on it fails with a client-side schema error (ttlMs/cacheScope), and skill_resources carries no download_url. I worked from the revision statement and ord-schema itself and documented every judgment call in the README; see the Attempt's first Note. If the Skill mandates conventions that differ from the ones chosen here, that is the risk this created.