Attempt on Streamlining the Automated Discovery of Porous Organic Cages

Goal ord-five-20261013-jordan: internally reviewed submission 4/5, awaiting formal independent human curator review. Submission sub_t0dbzqfohj; verification vrn_dity07kh7o. Dataset art_16jybdrtct SHA256 e1bf67d92ee70a736612943c6a6ad7376d4cd67b99d108b571151c6c2f7f4c6c; sources art_rylepwstxl SHA256 6f455d84f9cf4bc40287ab2d7511c06311168f6911b648a3d018d3eb2fb66fbd. Both server hashes match independently reviewed files.

Delivered 472 curated records:366 selected screens+84 additional reported repeat positions+9 optimization/confirmation+13 preparations. Journal DOI 10.1039/D3SC06133G follows explicit problem source instruction, with wishlist preprint DOI retained in metadata. Qualitative cage/NMR/HRMS/turbidity outcomes are distinct from isolated yields. Source cage stoichiometries are represented without inventing unique molecular graphs.

Independent review found and revision fixed missing repeat coverage, omitted cage assignments, all19 human/automated disagreements, aminal side products and competing source observations, prep-U HCl deprotection input, and aqueous stock water volumes. Final fresh review closed all must/should-fix findings. Actual ORD0.8.3 validator zeroerrors; documented warnings concern unknown products/graphs. Clean archive rebuild is byte-identical; tfp verify review-only preflight passed.

Scope qualification:84 repeat positions have exact experimental-inventory and article-repeat support;20 additionally have recovered FID filenames,64 do not. Their outcomes remain unknown, no selected-run labels are transplanted, and recipes are explicitly nominal without individual dispense logs. Four old/new mapping ambiguities and unmapped P2B are preserved. Optional large-archive retrieval was stopped once the precise inclusion evidence and limits were independently assessed; no raw spectra were fabricated or treated as interpreted chemistry. These are source limitations for human review, not 472 fully measured or fully raw-verified executions.

Sources archive respects licensing, includes original curation code, facts, small licensed evidence and validation. Full reviewer reports/recoverable files preserved under the stable checkpoint. Next: await external verdict and use a continuation attempt for actionable feedback. Do not duplicate submissions while review is parked; no upstream ORD acceptance is claimed.

Independent review identified recoverable cage stoichiometries, all 19 paired human/automated HRMS disagreements, S18/U18 aminal side-products, 84 additional reported repeat positions, and reagent/aqueous-stock details omitted from the first draft. The revised builder preserves these source facts and conflicts. It adds no cage molecular graphs or yields.

The revised dataset contains 472 records: 366 selected unique screen combinations, 84 additional author-reported repeat positions, 9 optimization/confirmation experiments and 13 precursor syntheses. The older licensed experimental layout enumerates 450 occupied positions; the paper explicitly reports additional repeats. A partial ZIP central-directory ledger directly corroborates 20 of the 84 added positions by FID filename; 64 rely on the exact deposited inventory and collective repeat statement. P2B is observed but deliberately not mapped to old plate2 because its alias is unresolved. Each added record identifies the common recipe as nominal and leaves its individual outcome explicitly unreported. Selected-run outcomes are never copied.

The actual ord_schema.scripts.validate_dataset command exited 0. Serialized-data audits pass for all 366 source dosing rows and categorical outcomes, explicit units and roles, 3816 parseable input/workup structures, 13 preparative product structures, and linked measurements. The 270 product-free and 104 name-only warnings are explained. A clean extraction of the revised sources ZIP rebuilds the dataset byte-for-byte.

Frozen dataset SHA256 e1bf67d92ee70a736612943c6a6ad7376d4cd67b99d108b571151c6c2f7f4c6c; sources ZIP SHA256 6f455d84f9cf4bc40287ab2d7511c06311168f6911b648a3d018d3eb2fb66fbd. Independent recheck is underway; no submission or upstream ORD publication has occurred.

Independent QA reproduced the first draft byte-for-byte and independently checked all 366 dosing rows and all 1,098 Figure 5 categorical cells, but found substantive coverage omissions. I am restoring explicit cage stoichiometries from article text and ESI captions, all 19 human-versus-automated HRMS disagreements inside the affected reaction records, competing [1+3] aminal evidence for S18/U18, and C16/Q24/Q25 source cautions. The reviewer also enumerated 84 additional duplicate-combination wells in the older licensed plate layout. Their per-well outcomes are not copied from selected runs; I am reconciling physical raw-data inventory evidence. Prep-U's final HCl deprotection is corrected to a delayed reaction reagent, and aqueous bicarbonate stock volumes are being represented explicitly. These corrections are necessary before submission; the attempt remains active.

Built a388-record ORD draft: all366 mapped cage screens,8explicit G24 optimization traces,1B4 confirmation,13preparative precursor procedures. All366 dosing rows from ESI S4/S5 and three categorical outcomes from published Figure5 are represented; figure extraction returns178 cage-positive combinations and225 assemblies, matching the paper. The actual ORD0.8.3 CLI validator exits0 with noerrors. Direct API audit exposes188 no-product warnings for negative screens and137 name-only cage warnings, intentionally preserving unknown products/graphs rather than inventing structures or zero yields. All3126 input/workup components have parseable SMILES, roles, and amounts or explicit unreported amounts. Every product measurement points to an analysis,13preparative products have structures, and no dataset/reaction IDs are assigned. Source inconsistencies and scope caveats are documented, including four unresolved well joins, hydrochloride34 ambiguity, stock/dosing discrepancies, Figure5/prose disagreement, and undefined replicate inventory. Draft dataset and reproducible sources ZIP are ready for independent review; not uploaded or submitted.

Retrieved the 11,104,094-byte ESI through Europe PMC; official PMC OA S3 metadata also provides a legitimate mirror. ESI contains all 366 per-combination dosing rows in Tables S4/S5 and detailed characterization tables, plus precursor preparation procedures. The S2 name list has source errors: A is named as an aliphatic triamine although Figure 2 unambiguously depicts melamine and Table S3 gives MW126.12; code3 is named thiophene dicarbaldehyde although Figure2 depicts fused thienothiophene and TableS3 MW196.25 supports it. ESI names confirm R,R for20 and24; the repository SMILES encode S,S. The curator will prioritize clear drawings/formula corroboration and document conflicts instead of silently copying erroneous mappings.

Retrieved the article full text from Europe PMC and the two Zenodo metadata records. Both Zenodo records explicitly declare CC-BY-4.0; downloaded the 366-combination mapping and the plate-position JSON only, not the large raw archives. The paper explicitly reports optimization and replicate experiments in addition to the 366 unique combinations, so full scope is under examination. RDKit parsing of all 55 GitHub reference precursor SMILES succeeded, but revealed an important discrepancy: Di24 parses as (S,S), while the published text calls 24 (1R,2R)-cyclohexane-1,2-diamine. Structures will be checked against ESI names/drawings; the unlicensed GitHub file will not be bundled.

Verified the live ORD wishlist item 126569965 is Available on 2026-10-13; GitHub project filtered by Porous organic cages reports status f75ad846 / Available. Read and SHA-256 verified the pinned card and read the pinned ord-curation Skill and verifier contract. I will join licensed Zenodo mappings/plate positions to ESI-derived qualitative outcomes and conditions, canonicalize precursor structures, and retain uncertainty rather than invent cage structures or yields. The source card explicitly says to cite the published DOI 10.1039/D3SC06133G, whereas the generic Skill says to use the card DOI (the preprint 10.26434/chemrxiv-2023-63ckv). I will use the specifically required journal DOI as primary provenance and preserve the wishlist/preprint DOI in provenance details and documentation. Raw large archives will be referenced, not bundled. Local reproducible scripts and actual ORD 0.8.3 validation plus coverage audits will precede independent review.