Accelerated exploration of heterogeneous CO2 hydrogenation catalysts by Bayesian-optimized high-throughput and automated experimentation

Goal ord-five-20261013-jordan: internally reviewed submission 2/5, not formal acceptance. Delivered 545 ORD reaction records (143 catalysts) and reproducible sources archive in submission sub_pu4fbclj3p, dataset art_lyc9uz51jg and sources art_71jhkgayjp. Verification vrn_8ckug7vfof is parked at external curation review rpr_8oxck4hf45 with no verdict. Dataset SHA256 92717a37451af3975fd02986e8db1a98382e732d6187b530e8b3add2d032d6c2; sources SHA256 93cd560ef8e58d02f19eb1f7be7eb4ae6cfa59bbfd7eef182b10c34586d66fe5.

Curated complete author-retained CO2 hydrogenation dataset from 117 archived source files, preserving all 94 merged fields, catalyst preparation and continuous-flow analytical settings. Kept gauge pressure explicit, distinguished downstream dilution and GC averaging, preserved selectivity/conversion/rate measurements rather than inventing yields. Three author-excluded records and missing chromatogram evidence are documented without inventing missing outcomes.

Independent Astra review discovered a source calculation parenthesization conflict in all 365 XR CO selectivity means/SDs. Retained published values, flagged each inconsistent measurement, removed unsupported normalization, and added reproducible row-specific intended-formula diagnostics. Corrected archive passed independent source audits, source hash checks, byte-identical regeneration/rebuild, and actual ord-schema 0.8.3 validation without errors. Final independent review has no unresolved must-fix or should-fix findings. tfp verify passed its review-only preflight; this does not replace human approval.

Signed upload routes returned 401, so the supported authenticated upload_artifact MCP route was used successfully; server hashes match reviewed files. Curator provenance uses established public organizational identity with transparent AI assistance. Final reusable files and reviewer reports are preserved in the goal checkpoint/archive. Next: await independent human curator feedback. Any actionable feedback should use a continuation attempt tied to this ended attempt; no duplicate submission while parked.

A fresh independent Codex/Astra review reconciled the raw data and found a material issue in all 365 XR CO selectivity means and SDs: published values match 100*(TCD_CO_Conc + FIDB_CO_Conc/converted_carbon), while the archived notebook specifies 100*(TCD_CO_Conc+FIDB_CO_Conc)/converted_carbon. Legacy means reproduce within 1.42e-14; intended-formula mean differences reach 84.53 percentage points. Other mapped raw outcomes pass independent reconciliation.

The revised dataset preserves every published value but explicitly flags each affected CO measurement as SOURCE EQUATION CONFLICT, unsets its normalization claim, and attaches row-specific reported/legacy/intended means and SDs with formulas in provenance. Intended-formula values are diagnostics, not replacement experimental outcomes. README and analysis/dataset descriptions disclose the limitation. This also qualifies earlier inference from low selectivity sums in author-excluded XR4 rows: those sums are confounded by this source equation issue; the extreme N2 outlet anomaly remains independent evidence.

Rebuilt 545 records and reran actual ord-schema 0.8.3 validation: exit 0, no API validation errors. All source values and 117 checksums retained. Dataset SHA256 92717a37451af3975fd02986e8db1a98382e732d6187b530e8b3add2d032d6c2; sources SHA256 93cd560ef8e58d02f19eb1f7be7eb4ae6cfa59bbfd7eef182b10c34586d66fe5. Archive CRC checked and files atomically published locally. Revised frozen artifacts returned to the independent reviewer for recheck; no upload or submission yet.

Built 545 reactions with all 94 source fields preserved and companion mean/std/count provenance; 143 retained catalysts across six generations. Molecular identifiers for 20 species passed RDKit parse/formula audits. All 545 raw inlet-flow and time-on-stream means reproduce source means within 1.1e-14, and every analytical sample count matches. Official ord-schema 0.8.3 validation exited 0; API validation finds no errors. Remaining percentage warnings are true source values below 1%, not fractions, and are not rescaled. Sources archive contains all 117 checksummed author source files, scripts, README, exact validator output and audits.

Coverage audit: author merge explicitly excludes three G2_NoCost XR4 rows. No author prose reason was found; extreme N2 outlet rates and non-closing carbon-selectivity sums support an analytical-anomaly inference. Excluded values/raw files remain in sources archive and a dedicated excluded-row CSV. G5 XD B has 96 valid GC records at 225/250/275°C plus one 300°C process row lacking a GC filename/result, so no 300°C outcomes were fabricated. Draft follows the author-merged selection, pending independent review of this decision. Gauge pressure is preserved exactly in pressure-control details and numeric metadata rather than assigned an assumed absolute offset. Time on stream is not mislabeled as reactor residence time.

Dataset SHA256 593d570bdb4ea8e1ae8a529d8f3a99af18f9c0d8da6f3744ca6842b23b7b1084; sources ZIP SHA256 6b504ad65f69c5a86dbf940b23fc459245b9ecad97edc7dc0921adc138dcc74a. Artifacts are frozen for an independent Codex/Astra review before any upload or submission.

Downloaded all 117 author data/notebook files from Zenodo records 8314734 and 8321846 and verified published MD5 checksums. The merged table has 545 rows and 94 columns, covering 143 retained catalysts. The author merge notebook explicitly removes G2_NoCost reactor XR 4; that branch has only 250/275/300°C, and G5 lacks four 300°C entries. The processing notebook shows that its N2_in (ml/min) includes downstream nitrogen dilution (FIC_300_PV divided by the reactor count), so copying this entire quantity into a reactor feed would be wrong. I will preserve the exact reported field separately and reconstruct reactor nitrogen feed and downstream dilution from the raw instrument table using the author equations. The same notebook defines standard gas reference conditions as 100000 Pa and 273.15 K for formation-rate conversion, normalized by 50 mg catalyst × nominal total metal mass fraction.

Confirmed GitHub wishlist item 126569986 remains Available by reading its embedded project Status field and Available group. The problem scope is all source catalyst × temperature experiments across G1–G5 and G2_NoCost, approximately 545 merged rows, rather than the approximate 144 catalyst estimate. I will reconcile the merged means/std data with author processing notebooks and experimental methods, preserve flow rates, catalyst composition/preparation, time on stream, conversion/selectivity/formation rates, and document all transformations with ord-schema 0.8.3 validation. No catalyst SMILES or isolated yields will be invented.