How to Triangulate Conflicting Sources in Client Research

Conflicting sources are normal in client research. The failure mode is not disagreement - it is quietly picking one source, averaging vibes, or burying the conflict so the recommendation looks cleaner than the evidence.

The unit of work is still a file: load-bearing claims triangulated with at least two independent, openable evidence lines, disagreements stated in the open, and a decision that survives hostile review (The Artifact Standard; The Client-Ready Research Report).

What triangulation means in practice

Triangulation is not "I read a lot." It is a bar for claims that carry weight.

• Two independent lines. Before a load-bearing claim ships, require at least two evidence lines that do not share the same upstream. A press release plus a blog that quotes the press release is one line, not two.
• Primary over secondary. Prefer filings, standards, official stats, product docs, named datasets, and dated interviews. Secondary summaries are bridges; they are not enough alone.
• As-of dates. Record when each source was accessed and, where it matters, the period the evidence covers. Stale agreement is not triangulation.

If you cannot open the source, you cannot claim it.

Claim → evidence A → evidence B → inference

Use a pattern a reviewer can audit without you in the room:

1. Claim in one sentence - the fact or finding you want to stand on.
2. Evidence A - primary-ish source, title, publisher, openable URL, as-of date.
3. Evidence B - a second independent line that supports, qualifies, or challenges A.
4. Inference - what you conclude for the decision, labeled as judgment, not as raw fact.

Keep claim, evidence, and inference in separate sentences. Mixing them is how disagreements disappear.

When sources disagree

Do not hide the conflict. Do one of three things, in writing:

1. Name the conflict. State what A says, what B says, and where they diverge.
2. Pick a stance with a reason. Prefer the more primary, more recent, more methodologically transparent, or more decision-relevant line - and say why.
3. Demote to Limits. If you cannot resolve it, move the claim out of Findings. "Unresolved: X vs Y as of [date]" is stronger than a fake consensus.

Averaging vibes ("somewhere in the middle") is not a method. It is how weak reports invent certainty.

What does not count as triangulation

• Secondary summaries alone (roundups, undated listicles, uncited model answers)
• Model-only claims with no openable source behind them
• Unverified screenshots, private Slack dumps, or "I saw it somewhere"
• Two pages that clearly copy the same upstream release
• A single strong source restated twice in different words

Those can appear in notes. They cannot carry a load-bearing claim in the file you send.

Keep a source log

Alongside the narrative, keep a short log the client can scan:

• Source title and publisher
• Openable URL (or explicit "not shareable / behind login")
• Access date and relevant as-of window
• Role: supports claim / challenges claim / context only

This is the same discipline as transparent evidence reporting in standards work and systematic review practice: show what you opened, when, and what it was used for - not a fake aura of completeness.

Checklist before you send

• [ ] Every load-bearing claim has at least two independent, openable evidence lines
• [ ] Claim, evidence, and inference are separated in the prose
• [ ] Conflicts are named; stance has a reason, or the claim sits in Limits
• [ ] Source log has URLs (or access limits) and as-of dates
• [ ] Non-counting evidence is demoted, not dressed up as triangulation
• [ ] The file ends with a decision-grade recommendation a stranger can act on

Conclusion

Triangulation is how research stays honest under pressure. Conflicting sources are expected; silent selection and vibe-averaging are not. Put the conflict in the file, require two independent lines for what matters, and ship a recommendation that can take a hostile question.

If you want that file to leave the chat as a workspace document, start free on mmarw.com. Live plans: Free, Starter $6, Professional $24.

Sources

• National Institute of Standards and Technology (NIST), AI Risk Management Framework - evidence, documentation, and risk communication discipline
• OECD, AI in Science - transparency and reproducibility expectations for AI-assisted research workflows
• Cochrane / PRISMA-style reporting ideas - state what was searched, included, excluded, and unresolved
• MMARW, The Artifact Standard
• MMARW, The Client-Ready Research Report
• MMARW, How to Write a Research Brief for AI