Technique
Consider-the-opposite is a directed search instruction. It asks the analyst to imagine that the favored conclusion is false and identify plausible reasons. It is more specific than “be fair” and does not require giving unsupported alternatives equal weight.
Generate one or more genuinely plausible contrary pathways. Separate the alternative hypothesis from evidence that it occurred. A failure story is not evidence of failure. Check whether your original model omitted a common cause, used a selected reference class, or treated intention as capability.
Diagnostic evidence
For each proposed challenge record:
- claim and its evidence origin;
- observable that would support the favored explanation;
- observable that would support the alternative;
- outcome that would remain ambiguous;
- expected forecast movement if the evidence is confirmed.
Prioritize tests with differing likelihoods under competing explanations. An executive’s repeated promise may be compatible with both delivery success and delay. A completed integration test independently witnessed before the deadline may be more diagnostic.
Use the same evidentiary standard on both sides. Do not dismiss unfavorable information as anecdotal while accepting favorable information of the same quality. Trace citations to originals and note when several stories share a source.
Crux analysis
A crux is a belief whose revision would substantially alter the conclusion. It need not be a single causal bottleneck. Estimate the forecast under alternative plausible crux values and show the difference.
Do not confuse disagreement reduction with accuracy. Two analysts can reach consensus by copying each other. Nor does a stubborn minority automatically possess privileged information. Preserve unresolved differences when the evidence does not decide them.
A blinded pass by another analyst may help when separately authorized. If only one agent is available, label it self-critique. Different personas or prompts from the same model are not independent human experts.
Stop conditions
Stop when the highest-impact contrary pathway has been examined to the depth the task justifies, when new searches repeat known evidence, or when access limits prevent further verification. Report residual uncertainty rather than inventing a balancing fact. Do not spend unlimited resources trying to prove every imaginable alternative false.
Output
Use a short claim-versus-challenge table only when it improves clarity. Include what changed and what did not. A probability may legitimately remain unchanged after review; this requires evidence-based explanation rather than mechanical movement toward 50%.