A bibliography should lead back to the evidence

Give each source a job

A reference list is useful when a reader can retrace a claim. It becomes decorative when every source seems to support the whole article and none points to the important detail. Begin with the sentence you want to substantiate. Then identify the work, the relevant location within it and what that material actually establishes.

DataCite's guidance treats data as citable research outputs, alongside familiar publications. Citing a dataset can make the underlying material easier to locate, inspect and reuse, while giving its creators credit. A paper describing a dataset and the dataset itself are related sources with different jobs.

Sources: s1

A durable identifier helps with the address

The DOI Foundation describes a DOI as a persistent identifier for an object or information about it. That helps distinguish the thing being cited from a temporary web address. It does not eliminate the need to open the resource and check that it supports the sentence beside the citation.

Sources: s2

An annotation worth keeping

For each important entry, add a short note: 'supports the sampling dates and method; does not establish a causal effect.' Keep the publication date, version, page or table reference and access date where relevant. If a source merely repeats another report, follow the chain before treating the repetition as independent confirmation.

For example, an invented claim that an island supported 10,000 inhabitants needs evidence about population. A reference proving that the island was once above water does not supply that evidence. The bibliography should reveal this gap, not conceal it behind an impressive-looking citation.