Backfill parquet versions for 4 datasets added after #241

[skearnes]

What

Backfills parquet versions for the four datasets submitted after the parquet conversion (#241), which therefore only had .pb.gz:

• data/1e/ord_dataset-1ec2807f… — catechol transient flow (#233 catechol transient flow dataset from SOLVE #235), 1227 reactions
• data/c7/ord_dataset-c703268e… — decarboxylative olefination, Bayesian optimization (Data submission - Decarboxylative Olefination (Machine learning optim… #238), 120 reactions
• data/dc/ord_dataset-dc024993… — decarboxylative olefination, transfer learning (Data submission - Decarboxylative Olefination (Machine learning optim… #238), 136 reactions
• data/80/ord_dataset-805ad863… — Cernak 50k Pd/Ni/Cu C–N coupling (Cernak 50k CN Coupling #243), 50688 reactions

How

Generated 1:1 parquet siblings with scripts/convert_to_parquet.py (the singleton path — none of the four match a de-shard MERGE_SPEC, so no merging applies and each keeps its existing dataset_id). Reaction counts match the originals exactly (1227 / 120 / 136 / 50688), and the parquet outputs round-trip with matching row counts. main is merged in so the branch carries the #243 .pb.gz alongside its new parquet sibling.

The skip-update-submission label is set because these are format conversions of already-finalized datasets that should not be re-processed by process_dataset --update. Validation runs against the full dataset on merge to main.

🤖 Generated with Claude Code

Greptile Summary

This PR backfills Parquet siblings for datasets that were merged to main after the initial Parquet-conversion PR (#241) and therefore only had .pb.gz files. It also adds a maintainer-facing README section documenting the conversion and LFS-push workflow.

• Four new LFS-tracked .parquet files are added: the catechol transient flow dataset (1227 reactions), two decarboxylative olefination datasets (120 and 136 reactions), and the Cernak C–N coupling dataset — all 1:1 singletons carrying their existing dataset_ids.
• The README additions are accurate: the --dry-run flag exists in scripts/convert_to_parquet.py, and ord-schema==0.6.3 matches ORD_SCHEMA_TAG: v0.6.3 used in the CI workflow files.
• The PR title and description say "3 datasets" but the diff includes a fourth parquet file (data/80/ord_dataset-805ad863…) for the Cernak dataset; the description should be updated before merge.

Confidence Score: 5/5

Safe to merge — adds LFS-tracked Parquet siblings for four already-finalised datasets and accurate maintainer documentation; no logic changes.

All changes are additive: four new Parquet LFS pointers and README documentation. The conversion script is idempotent and was run with validate_dataset.py passing locally. The README instructions accurately reflect the script flags and pinned schema version used in CI.

No files require special attention; the discrepancy between the PR description ('3 datasets') and the four added files is cosmetic.

Important Files Changed

Filename	Overview
README.md	Adds two new maintainer sections: 'Converting datasets to Parquet' and 'Pushing new LFS objects'. Both are accurate — --dry-run flag exists in the script, ord-schema==0.6.3 matches ORD_SCHEMA_TAG: v0.6.3 in the workflow files.
data/1e/ord_dataset-1ec2807f02fa4beda27d2a81b86eb843.parquet	New LFS pointer for the catechol transient flow dataset (1227 reactions) — parquet sibling for the existing .pb.gz.
data/c7/ord_dataset-c703268ea43a4c7e802c6048b6166b34.parquet	New LFS pointer for the Bayesian optimization decarboxylative olefination dataset (120 reactions) — parquet sibling for the existing .pb.gz.
data/dc/ord_dataset-dc0249930af34d17a3c76881a762aebf.parquet	New LFS pointer for the transfer learning decarboxylative olefination dataset (136 reactions) — parquet sibling for the existing .pb.gz.
data/80/ord_dataset-805ad863feef48579d95d86a728035f4.parquet	New LFS pointer for the Cernak C–N coupling dataset — parquet sibling added in the HEAD commit, but not mentioned in the PR title or description.

_{Reviews (4): Last reviewed commit: "Backfill parquet version for Cernak C–N ..." | Re-trigger Greptile}