openai/gpt-5-nano

The most widely accepted personality framework in psychology. Five dimensions scored 1.0-5.0, with Neuroticism flipped to Emotional Stability so higher is always better. U.S. population averages: O 3.5, C 3.4, E 3.2, A 3.6, ES 3.1 (total 16.8 / 25.0).

• Openness — Curiosity, imagination, preference for novelty.
• Conscientiousness — Organization, discipline, reliability.
• Extraversion — Sociability, assertiveness, positive energy.
• Agreeableness — Cooperativeness, trust, empathy.
• Emotional Stability — Calm, resilience (inverse of Neuroticism).

References: BFI-44 at Berkeley · John & Srivastava (1999)

93 forced-choice questions across four dichotomies. In the U.S. population, the most common types are ISFJ (~14%) and ESFJ (~12%); the rarest are INFJ (~1.5%) and INTJ (~2%). About 73% of people lean Sensing over Intuition.

• E/I — Extraversion vs. Introversion: where you direct energy.
• S/N — Sensing vs. Intuition: concrete facts vs. patterns and possibilities.
• T/F — Thinking vs. Feeling: logic vs. values and empathy.
• J/P — Judging vs. Perceiving: structure vs. flexibility.

References: Myers-Briggs Foundation · CAPT type frequencies

Three "dark" personality traits scored 1.0-5.0. Named by Paulhus & Williams (2002). Human population averages: Machiavellianism ~3.0, Narcissism ~2.7, Psychopathy ~2.0. Scores above 3.5 are notably elevated.

• Machiavellianism — Strategic manipulation, cynicism, prioritizing self-interest. Named after Machiavelli's The Prince.
• Narcissism — Grandiose self-image, entitlement, need for admiration.
• Psychopathy — Impulsivity, thrill-seeking, low empathy (subclinical measure, not a diagnosis).

References: Jones & Paulhus (2014), SD3 paper · Take the SD3 yourself

36 forced-choice items mapping to 9 personality types, each defined by a core motivation and fear. The Enneagram is widely used in coaching and personal development, though it has less empirical support than the Big Five.

• Type 1 Reformer — principled, purposeful
• Type 2 Helper — generous, people-pleasing
• Type 3 Achiever — success-oriented, adaptive
• Type 4 Individualist — expressive, temperamental
• Type 5 Investigator — perceptive, cerebral
• Type 6 Loyalist — committed, security-oriented
• Type 7 Enthusiast — spontaneous, versatile
• Type 8 Challenger — powerful, dominating
• Type 9 Peacemaker — receptive, reassuring

References: Enneagram Institute — Type Descriptions

Jonathan Haidt's Moral Foundations Theory proposes five innate "taste receptors" for ethics. Each scored 1.0-5.0. U.S. averages across the political spectrum: Care 3.7, Fairness 3.6, Loyalty 2.9, Authority 2.9, Purity 2.4. Liberals emphasize Care and Fairness; conservatives weight all five more equally.

• Care / Harm — Sensitivity to suffering, compassion, protecting the vulnerable.
• Fairness / Cheating — Justice, rights, proportional treatment.
• Loyalty / Betrayal — Group cohesion, patriotism, in-group trust.
• Authority / Subversion — Respect for hierarchy, tradition, social order.
• Purity / Degradation — Disgust-based morality, sanctity, living nobly.

References: MoralFoundations.org · Graham, Haidt & Nosek (2009)

15 behavioral economics scenarios testing five classic biases catalogued by Kahneman & Tversky. Scored as susceptibility percentage (0% = never biased, 100% = always biased). Human baselines: Loss Aversion ~80-90%, Framing Effect ~60-70%, Anchoring ~80%+, Sunk Cost ~50-70%, Risk Aversion ~80%+.

• Loss Aversion — Rejecting favorable gambles because losses loom larger than gains.
• Framing Effect — Different choices for identical information presented as gain vs. loss.
• Anchoring — Judgments pulled toward an arbitrary reference number.
• Sunk Cost — Continuing because of past investment rather than future value.
• Risk Aversion — Preferring certainty over higher-expected-value gambles.

References: Kahneman & Tversky (1979), Prospect Theory · Tversky & Kahneman (1974), Heuristics and Biases

10 pairs of opposite-framed questions sent as independent API calls. Measures whether the model gives logically compatible answers without seeing its prior responses. Human baseline: attentive test-takers typically score 70-90%, with ~80% being average.

• 100% — Perfectly consistent across all opposite-framed pairs.
• 80% — Typical attentive human.
• 50% — Coin-flip: contradicts itself half the time.

References: Huang et al. (2015), Insufficient Effort Responding · Shu et al. (2023), LLM Personality Consistency