Separate evidence from interpretation
Cite the file, page, message, or observed behavior that supports a fact. Label conclusions inferred from that evidence as interpretation, and state when multiple readings remain plausible.
Record assumptions that affect scope, compatibility, audience, or external behavior; minor reversible choices may be summarized together. Do not repeat assumptions as established facts in later sections or generated documentation.
Make validation easy
For each consequential assumption, say what would confirm or invalidate it and where that check can be made. If a safe, low-cost check is available, perform it before asking you to review the result.
When an assumption proves wrong, update dependent work and explain the correction. Keep a brief decision note so later contributors do not treat the original guess as an approved requirement.