Reference section

Preserving a reference dataset release

A saved dataset release should identify the files, their checksums, the date obtained and what the records mean. Preserve those details together so a later reader can reproduce the analysis and distinguish an editorial revision from a new animal-health observation.

Updated 542 words · 3 min read

Work with the reference records

Save the files that belong together

Begin with the data download guide and save the files required for your task. Include the matching checksum list. Keep them in a clearly dated local folder, along with a short description of the question being investigated.

The public download addresses describe the current export. They should not be assumed to preserve every earlier version at the same URL. If an analysis must be reproducible, retain the exact copy used rather than relying on downloading it again months later.

Record the unit and scope

A source manuscript, a canonical reader page, a source document and a bibliography lead are different record types. A release note should say which one is being counted. Consolidation can reduce the number of reader pages while retaining all original manuscript IDs.

For example, a source-coverage table can begin from the article CSV and preserve every source identifier. A reading directory can group by canonical URL. Both can be correct while producing different totals. Explain the unit alongside the result.

What a checksum establishes

A checksum identifies the bytes in a particular file. Comparing a local SHA-256 value with the export's list can show whether the file matches that listing. It cannot establish scientific validity, regulatory currency or suitability for a new use.

Keep the checksum list itself with the saved release. If a later file has changed, compare the records and field meanings rather than assuming that a changed hash tells you what changed in the content.

Preserve the analysis steps

Record filters, joins, removed duplicates and any manual corrections. If a title is standardized or a bibliography record reconciled, keep the earlier value and the source for the correction. Another reader should be able to distinguish a change in the underlying export from a change introduced during analysis.

Where an exported record links to an external source, preserve the document identity and relevant date fields. The source can change independently of the site's export. The historical-record guide explains how publication dates, event dates and later revisions fit together.

Current public coverage

The present exports account for the supplied manuscript collection and describe its consolidated reading destinations, reference metadata, sources and bibliography leads. The site does not present an invented back-catalog of dated dataset releases. Earlier local build packages are separate working artifacts rather than additional public disease observations.

How should a correction be reported?

Identify the file, record ID, field and proposed source-supported correction through corrections. Include the saved file's date and checksum when available so the exact version can be identified.

Add a release manifest

A publisher can keep a manifest with the release ID, schema version, filenames, SHA-256 values, input identities and export settings. Preserve the manifest with the exact files. For a deterministic export, compare two runs from identical preserved inputs and settings; record expected nondeterministic fields separately. Matching outputs establish repeatability under those conditions, not scientific validity. If an input or schema changes, identify the result as a different release rather than overwriting the archived copy.