All numeric cases below are synthetic teaching examples, not empirical research findings.
Example: product confidence
Initial estimate: 75% chance of delivery. Favored explanation: the feature demo looks complete.
Opposite explanation: the demo is a scripted path and integration work remains. Diagnostic evidence: test coverage on actual dependencies, unresolved blockers, and a date-stamped release candidate. “The team seems confident” distinguishes little.
If no integration evidence is available, retain a provisional forecast or widen the assumption sensitivity; do not fabricate a test failure.
Example: no automatic discount
An adversarial review finds that all supposed contrary evidence is copied from a disproven report. Record that finding. Do not subtract 10 percentage points merely to appear balanced.
Counterexample
Giving a fringe explanation 50% because “both sides deserve a hearing” is false balance, not considering the opposite.