Scaffolding the Strategist: Architecture-Dependent Reasoning Interventions in Hotelling Spatial Markets
Quick Answer
This study evaluates the impact of structured reasoning interventions on GPT-4.1-mini and GPT-5-mini in Hotelling spatial markets.
Quick Take
This study evaluates the impact of structured reasoning interventions on GPT-4.1-mini and GPT-5-mini in Hotelling spatial markets. Results show that commitment scaffolding enhances the standard model's performance, while degrading the reasoning-optimized model, highlighting architecture-dependent effects on strategic reasoning.
Key Points
- Commitment scaffolding improves GPT-4.1-mini by +0.21 but degrades GPT-5-mini by -0.63.
- Principled separation shows opposite effects: +0.31 for GPT-5-mini and -0.40 for GPT-4.1-mini.
- Adversarial stress-testing harms both models, with a 2.6x greater degradation for the reasoning model.
- A persistent declarative-procedural gap exists, with separation closing this gap for the reasoning model.
- The study involved 720 responses across eight questions, demonstrating significant architectural interactions.
Paper Resources
📖 Reader Mode
~2 min readAbstract:We investigate whether structured reasoning interventions improve the strategic economic reasoning of large language models, and whether their effects depend on model architecture. Using Hotelling's linear city model as a diagnostic vehicle, we evaluate GPT-4.1-mini (a standard instruction-following model) and GPT-5-mini (a reasoning-optimized model) under five conditions - an unscaffolded baseline and four reasoning interventions - across eight questions spanning deductive and abductive reasoning, three prompt framings, and three repetitions per condition, yielding 720 individually judged responses. We find a statistically significant crossover interaction between scaffolding type and model architecture ($t(7) = 4.79$, $p = 0.002$, $d = 1.69$): commitment scaffolding improves the standard model ($+0.21$) while degrading the reasoning model ($-0.63$), and principled separation shows the opposite pattern ($-0.40$ vs. $+0.31$). Both crossovers are individually significant (commitment: $p = 0.040$; separation: $p = 0.002$) and hold across all eight questions with 7/8 directional consistency. Adversarial stress-testing harms both models, with $2.6\times$ greater degradation for the reasoning model ($-1.47$ vs. $-0.57$; $p = 0.038$), and the damage correlates negatively with baseline difficulty ($R^2 = 0.36$, $p = 0.014$). We further document a persistent declarative-procedural gap in which both models identify correct strategies at rates far exceeding their ability to execute them; separation fully closes this gap for the reasoning model while no intervention helps the standard model.
| Comments: | 26 pages (11 main + 15 appendix), 6 figures, 4 tables. Accepted at the ICLR 2026 Workshop on LLM Reasoning |
| Subjects: | Artificial Intelligence (cs.AI); Computer Science and Game Theory (cs.GT) |
| MSC classes: | 91A80, 68T50 |
| ACM classes: | I.2.7; J.4 |
| Cite as: | arXiv:2607.09743 [cs.AI] |
| (or arXiv:2607.09743v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.09743 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Pratyush Singh [view email]
[v1]
Fri, 3 Jul 2026 10:23:31 UTC (811 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Adversarial Social Epistemology for Assemblies of Humans and Large Language Models
The paper introduces Adversarial Social Epistemology (ASE) to analyze how agents manipulate trust in public communications, highlighting mechanisms that undermine the reliability of testimony and inference. It critiques existing frameworks like epistemic bubbles and misinformation diffusion, proposing a new language for understanding trust breaches and auditing inferential chains in densely interactive environments involving humans and large language models.