How to Run a Truth-Seeking Review Group

How to Run a Truth-Seeking Review Group

A truth-seeking group should make it safer to disagree, harder to hide weak evidence, and normal to change the conclusion after the meeting. It is not a debate club, friendship test, therapy group, legal process, or substitute for professional review.

Establish the invitation

Name the question, decision owner, participants, purpose, confidentiality boundary, record policy, and time box before the meeting. Tell people whether the group is advising, deciding, or reviewing.

Participation should be informed and voluntary where the setting permits it. A friend may not want every casual conversation converted into performance analysis. A direct report may not feel free to decline a leader’s “optional” truth session.

Amy Edmondson’s Harvard Business School paper on psychological safety examines interpersonal risk and learning in teams. The analysis is not a guarantee. The group must still account for power, retaliation, status, identity, and consequences.

Prepare the claim and its rival

Write the focal claim in a form that could be wrong. Then write at least one competing explanation before asking the group.

For example, “this product is not growing because distribution is weak” should be paired with plausible alternatives involving value, audience, price, trust, execution, measurement, or timing.

flowchart LR
    A["Question and competing claims"] --> B["Source and context check"]
    B --> C["Independent views"]
    C --> D["Evidence comparison"]
    D --> E["Confidence update"]
    E --> F["Decision and correction trigger"]

The rival explanation prevents a meeting from becoming a search for support.

Inspect sources laterally

For each material source, ask who produced it, for what purpose, from which evidence, and with which incentives or limitations.

Stanford’s lateral-reading curriculum recommends leaving a site to see what independent sources say about it. Apply that before spending the meeting interpreting a polished but unreliable page.

Separate direct evidence, credible secondary reporting, expert judgment, anecdote, forecast, and personal preference. None of those categories is useless, but they should not carry the same weight.

Collect views before discussion

Ask each participant to record an initial conclusion, confidence, key evidence, uncertainty, and disconfirming signal before the group conversation.

Independent entries reduce the chance that the first confident speaker determines the room. They also create evidence of whether the discussion changed anyone’s view.

The facilitator should invite the least powerful or most dissenting view without forcing a person to represent an entire identity or disclose sensitive information.

Compare evidence, not personalities

Discuss the strongest support for each explanation. Ask which facts are shared, which are disputed, which sources are missing, and which result would distinguish the claims.

Assuming good intent can lower unnecessary defensiveness, but it is not a conclusion about behavior. Conflicts of interest, repeated harm, policy violations, manipulation, and bad evidence still require direct treatment.

Gary Klein’s premortem method can help surface concerns that people are reluctant to state. Imagine the chosen plan failed and ask each person for plausible causes before debating them.

Record a confidence update

At the end, each participant should record the most likely explanation, confidence, important uncertainty, and evidence that would change the view.

The decision owner then records the decision, reason, dissent, action, evidence to collect, review date, and correction trigger. Consensus is not required unless the actual governance or policy process requires it.

For higher-risk matters, use the right authority. A peer group cannot decide a medical diagnosis, legal right, employment action, security exception, financial recommendation, or board obligation merely because the meeting felt rigorous.

Correct the record

Truth-seeking continues after the meeting. At the review date, compare the expected signal with what happened. Update the conclusion and link the new evidence.

NIST’s AI Risk Management Framework was designed for AI risk, not personal peer groups. Its separation of Govern, Map, Measure, and Manage is still a useful reminder that context, measurement, oversight, and response are different jobs.

The group should also review itself. Did anyone withhold a view? Was confidentiality kept? Did the facilitator reward dissent or only tolerate it? Did the decision owner represent disagreement fairly?

Keep friendship and review distinct

Friends can be excellent reviewers, but the roles should not be silently merged. Ask for review explicitly, define when the session ends, and allow the relationship to return to ordinary life.

A truth-seeking group succeeds when participants can be wrong, revise, and remain respected. Its product is not bluntness. Its product is a better decision record and a credible route to correction.

About this guide

This protocol was developed from Venture Step E051 and cited research with AI assistance. It requires context-specific facilitation, confidentiality, policy, employment, legal, safety, and professional review when those concerns apply.