Activate this model when calibrating trust in a person (e.g., for partnership, delegation, advice, or any significant reliance).
Revealed behavior over time is the most predictive signal. Stated intentions are nearly worthless as standalone data.
This is not cynicism. It is the empirical finding that self-report and actual behavior are weakly correlated, especially under pressure, especially when interests diverge. Build your model of a person primarily from what they have done, not what they say they will do.
Track Record: Behavior over time, across multiple instances, especially under stress. Common error: Allowing one compelling explanation to reset the track record to zero
Incentive Mapping: What does this person gain from each possible action? Common error: People reliably follow incentives even when they sincerely believe they are following principles. Taking stated motivations at face value; ignoring structural incentives
Consistency: Does behavior cohere across different contexts, different audiences, different stakes? Common error: Only observing the person in one context (usually favorable to them)
Skin in the Game: Is this person exposed to the consequences of their own advice or recommendations? Common error: Weighting the advice of confident-sounding advisors who bear no downside
Variance Profile: Is this person reliably average (low variance) or occasionally brilliant and occasionally catastrophic (high variance)? Common error: Treating high-variance people as reliable or as unreliable, rather than as high-variance
High: Behavior under conditions where it was costly to behave well; when honesty was expensive
Medium: Consistent behavior across multiple low-stakes situations
Low: Single instances; behavior in favorable conditions; self-report
One instance of high-quality track record evidence (behaving well when it was costly) outweighs many instances of medium-quality evidence.
Halo effect: Positive impressions in one domain contaminating assessment in another
Narrative absorption: Being convinced by a compelling account of past behavior that resets your model
Context blindness: Judging behavior without accounting for the constraints the person was operating under