FAIR data practices

Datasets have stable UUIDs and version numbers. Each indexed reaction has a UUID resolving to its dataset version, file and source row. Reprocessing the same file preserves row identifiers; new versions and independent sources retain separate identities. Public release records expose JSON, CSL-JSON and Schema.org JSON-LD metadata.

Public data can be accessed through the website, HTTPS APIs and MCP. Protected content requires authorization. Minimal citation metadata is retained after public release, even if files are withdrawn or deleted. Never-public drafts and protected data are not made public by this mechanism.

The website, MCP and RxnSeek use the same contribution rules and field catalog. Structured records preserve chemical identifier types, reported quantities and units, measurement subjects and provenance. Human tables retain reported wording for parsing; complex experiments use the structured contract.

Version citations include authors, title, publication year, version and link. Datasets retain provenance and licensing; export manifests identify selected records and their sources. Author ORCIDs are optional. Citations identify a specific published version. Once DOI registration is connected, new content versions receive distinct version DOIs; bibliographic corrections retain the same DOI. An optional concept DOI can represent the version series. Registration is not connected yet.

Long-term preservation, domain-vocabulary alignment, external indexing and DOI registration require continuing work. Retained public metadata starts with this release and backfills currently accessible published versions. Historical private records are not treated as evidence of public access permission.