npbuilds Kahneman Framing — Cognitive Failure Modes in Problem Statements
The four failure modes binding-vow's audit subdomain checks for are not bugs in reasoning — they are features of how human cognition produces problem statements automatically. Understanding the mechanism makes detection routine; without the mechanism, detection feels like art.
The four modes are independent. A single problem statement can exhibit all of them simultaneously.
1. Framing Effects (Tversky & Kahneman)
The same factual content, presented in different framings, produces systematically different judgments and choices. Framing is not noise — it is a determinant of decision.
Canonical example (Asian disease problem, 1981):
- Frame A: "200 of 600 will be saved" → 72% choose certain
- Frame B: "400 of 600 will die" → 78% choose risky
The two frames are mathematically identical. The choices reverse.
| Framing axis | Effect |
|---|---|
| Gain vs loss | Loss frames produce risk-seeking; gain frames produce risk-aversion |
| Default vs opt-in | Default settings dominate even when costs are negligible |
| Reference point | Performance evaluated against a salient anchor; arbitrary anchors stick |
| Granularity | Same data presented per-event vs per-aggregate produces different judgments |
| Active vs passive voice | "Mistakes were made" vs "I made mistakes" allocates blame differently |
Implication for problem statements: before treating a stated problem as the problem, ask what frame is in use. The frame chooses what counts as evidence and what counts as a solution.
2. Attribute Substitution (Kahneman & Frederick)
When asked a hard question, the mind unconsciously substitutes an easier question and answers that one instead. The substitution is automatic and invisible.
| Hard question | Substituted (easier) question |
|---|---|
| "Is this candidate qualified for the job?" | "How likable is this candidate?" |
| "What is the probability this startup succeeds?" | "How vivid / coherent is the founder's pitch?" |
| "Should I take this medication?" | "How do I feel about the doctor?" |
| "Is this argument valid?" | "Do I agree with the conclusion?" |
| "What's the right answer?" | "What's the easy answer?" |
The substitution mechanism is universal and automatic (System 1, in Thinking, Fast and Slow). The detection mechanism is rare and effortful (System 2). Binding-vow's audit makes the detection routine by asking explicitly: what easier question might be answering instead?
3. Einstellung Effect (Luchins 1942)
Once a familiar problem-solving pattern has worked, it suppresses awareness of better alternatives. The pattern doesn't just compete with the alternative — it makes the alternative invisible.
Original water-jar studies (Luchins 1942): subjects taught a 5-step solution for measuring water volumes failed to see a 2-step solution that was available, even when the 2-step was the only valid solution for a later problem.
Modern instances:
| Domain | Einstellung pattern |
|---|---|
| Software | "We've always done it with a relational DB" — graph or document store invisible |
| Medicine | "The differential includes A, B, C" — D never considered because A-C are the trained set |
| Investing | "This looks like the 2008 setup" — current setup's distinguishing features ignored |
| Strategy | "Our playbook works" — opponent has read the playbook |
Implication for problem statements: when a stated problem fits a familiar pattern, the audit must check whether the pattern is suppressing better formulations. Pattern-matching produces the first viable framing, not the best.
4. Type III Error (Mitroff & Kilmann 1978)
A Type I error is a false positive (rejecting a true null). A Type II error is a false negative (failing to reject a false null). A Type III error is the right answer to the wrong question — a perfectly valid solution to a problem that wasn't the actual problem.
Type III errors are the most expensive cognitive failure mode in real-world work because:
- They look successful from inside the analysis (the math checks out)
- They require external feedback (something that should have happened didn't) to detect
- The cost is fully incurred before detection is possible
The XY problem (Raymond, Jargon File) is a special case: someone asks how to do X (their attempted solution) instead of asking about Y (their actual goal). All XY problems are Type III errors waiting to happen.
Diagnostic for Type III:
- Did anyone ever ask "is this the right question to be asking"?
- Could the proposed answer be correct AND the situation still be unimproved?
- Has the question been audited against the actual goal, or against a proxy goal?
How binding-vow uses these four modes
Each mode maps to a specific audit skill (Phase 6):
| Cognitive mode | Audit skill | Detection question |
|---|---|---|
| Framing effect | frame-rotator (Phase 4 reframing) + assumption-excavator (framing category) |
"What other frames exist? What does each make visible?" |
| Attribute substitution | answerability-tester + statement-grader (specificity axis) |
"What easier question might this statement be answering instead of the hard one?" |
| Einstellung | frame-rotator (Munger inversion) + inversion-tool |
"What pattern is suppressing alternatives here?" |
| Type III | xy-detector + statement-grader (root-vs-symptom axis) |
"What is the actual goal? Is this the question that gets us there?" |
The audit subdomain is Reiter-Palmon's "automatic problem construction" finding made operational. Construction is automatic; the audit makes the failure modes detectable.
Common Mistakes
- Treating framing effects as bias to be eliminated. They cannot be eliminated; only made explicit.
- Looking for attribute substitution only in others' reasoning. Yours has it too. Especially yours.
- Believing Einstellung only affects novices. Experts are more susceptible — the pattern library is denser.
- Confusing Type III with Type I/II. Type I and II are statistical; Type III is conceptual. Different machinery, different detection.
- Relying on intuition to detect any of these. The whole point is that intuition produces the failure mode in the first place.
Connections
xy-detector(binding-vow, future) — primary Type III detectionstatement-grader(binding-vow, future) — six-axis grader checks all four modes implicitlyanswerability-tester(binding-vow, future) — primary attribute-substitution detectionframe-rotator,inversion-tool(binding-vow, future) — primary framing-effect and Einstellung mitigationassumption-excavator(philosophy/logic) — surfaces framing assumptions explicitly
Sources
- Tversky, A., & Kahneman, D. (1981). "The Framing of Decisions and the Psychology of Choice." Science, 211(4481), 453–458.
- Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. (Comprehensive treatment of System 1 / System 2 and attribute substitution.)
- Kahneman, D., & Frederick, S. (2002). "Representativeness Revisited: Attribute Substitution in Intuitive Judgment." In Heuristics and Biases (Gilovich, Griffin, & Kahneman, eds.), Cambridge University Press.
- Luchins, A. S. (1942). "Mechanization in Problem Solving: The Effect of Einstellung." Psychological Monographs, 54(6), 1–95.
- Mitroff, I. I., & Kilmann, R. H. (1978). Methodological Approaches to Social Sciences. Jossey-Bass. (Type III error.)
- Raymond, E. S. (compiler). The Jargon File — entry on the XY problem. (Folk-canonical formulation.)