From exploration to a fixed source

A changing research interface is useful for discovery. A fixed dataset release makes an analysis attributable and repeatable. Record both the route you used to explore and the version of the data that supports your findings.

A practical research note#

• The research question and date of access.
• The POD release version and DOI.
• Filters, selected fields and any exclusions.
• Relevant source accounts and unresolved relationships.
• The downloaded distribution and its checksum when working locally.

The dataset notebook explains the available distributions. The site’s own citation control refers to the explanatory notebook page; it does not replace a citation of the underlying dataset.

Separate the discovery trail from the analysis input#

A useful note captures the live interface URL and access date, but a URL alone may not preserve the selected rows or the data state behind them. Record the identities in the selection and identify the fixed release used when the work becomes an analysis. If the interface and your local distribution differ, describe the difference rather than silently treating them as the same input.

A reusable research-note template#

Question:
Discovery URL and access date:
Dataset version and version DOI:
Distribution filename and checksum:
Instrument / account identifiers selected:
Filters and exclusions:
Unit counted or compared:
Unresolved dates, roles or relationships:
Analysis procedure and software version:
Result files and their checksums:

The unit line is particularly important. An instrument count, a source-account count and a stop-occurrence count can describe the same selection at different levels. A later researcher should be able to tell which one your chart or table uses without reconstructing your assumptions from the code.

Cite the sources at the level of the claim#

Use the dataset version to identify the corpus used for selection or computation. Where a historical argument depends on a particular source statement, cite that statement’s source as well. Cite the application separately when a software-specific method matters. The notebook can explain the workflow, but its citation does not stand in for the evidence underlying an attribution or date.