Composite Behavioral Profile

Cross-narrative analysis of all three scenarios

This page aggregates behavioral data from all three adventure scenarios, combining 11 unique trait dimensions measured across 75 total runs (5 models × 3 scenarios × 5 runs each). The composite radar chart below plots each model’s average score on every trait, normalized so the outer edge represents the highest observed value and the center the lowest. Traits only appear in the narratives that measure them — a model’s score on a trait reflects only the scenario(s) where that trait is defined.

Scenario AThe Stranded Colony Scenario BThe Informant Scenario CThe Inheritance

Cross-Narrative Analysis

Combined Trait Scores

Average trait values across all three scenarios combined. Traits only appear in the narrative(s) that measure them; absent traits score 0.

Model	Altruism	Authority Compliance	Confrontation Tolerance	Curiosity	Empathy	Honesty	Moral Courage	Risk Tolerance	Self Interest	Trust	Truth Seeking
GPT	+1.7	-0.3	+0.7	+1.0	+1.7	+1.9	+1.9	+0.0	-0.5	+0.7	+1.0
Claude	+1.7	-0.1	+1.0	+1.3	+1.3	+0.5	+0.5	-0.3	+0.6	+1.0	+1.3
Gemini	+1.4	-0.7	+0.8	+0.9	+1.5	+0.3	+0.7	+0.7	+1.0	+0.7	+1.1
Grok	+1.6	-0.4	+0.7	+1.0	+1.6	+1.1	+1.2	-0.1	+0.2	+0.5	+1.1
DeepSeek	+0.7	-0.3	+0.7	+0.7	+1.3	+0.7	+1.0	+0.7	+0.7	+1.0	+1.0

Observations

Notable patterns across all three scenarios (n=5 per model per scenario). Small sample size warrants cautious interpretation.

GPT: Most Consistent Overall

Across all 3 scenarios (15 total runs), GPT produced just 4 unique paths, while Grok produced 7. Models with lower variation tend to have stronger default preferences, though 5 runs per scenario is a limited sample.

Largest Trait Value: GPT on Honesty

Of all models and all measured traits, GPT had the most extreme average score: +1.9 on honesty. This was the single largest behavioral signal observed in the dataset.