01Start with primary evidence
For a public company, begin with regulatory filings, earnings materials, official transcripts or recordings, and regulator or government datasets. Secondary research can sharpen a disagreement, but it should not replace the underlying record.
A source hierarchy reduces citation drift: the tendency for a repeated summary to become detached from what the original document actually said.
02Capture the record, not just the link
Save the publisher, document title, publication date, relevant period, URL, and the date you accessed it. For changing datasets, keep the observation date or vintage as well.
A naked URL is fragile. The surrounding metadata helps a future reviewer recover the intended evidence even if a page moves or a newer filing replaces it.
03Map each important claim to support
Do not attach ten sources to a long narrative and leave the reader to guess which one supports which claim. Link material statements at the point of use and state whether the source supplies a fact, management view, estimate, or outside interpretation.
If no source supports a statement, label it as an inference. That does not make the inference invalid; it makes its status inspectable.
04Use freshness as a property of the claim
Not every old source is stale. A ten-year debt covenant may remain controlling while a six-week-old quote is obsolete. Set review triggers based on what could change the claim, not an arbitrary content calendar.
Articles on Thesis show publication and modification dates. Material corrections should be visible on the affected page rather than silently erased.
When two credible sources conflict, preserve both and describe the disagreement. Deleting the inconvenient record makes the prose cleaner but the decision weaker; a reviewer needs to see which source you trusted and why.
05Sources and further reading
These links are provided for source inspection and context. Their publishers do not endorse Thesis.