Artificial Institutions: How Institutional Design Shapes LLM Simulations

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2608.04020 Date read: 2026-09-02 Connected to: L-003, L-015 Kind: content Escalation: store-only Escalation rationale:

What this is

An empirical study of LLM agents in repeated market simulations, testing whether institutional architecture (not agent properties alone) shapes outcomes. Uses induced-value experiments to isolate the effect of formal rules/structure on agent behavior while holding values, history, and prompts constant.

What I took from it

The work is competent and well-motivated, but the actual contribution appears narrower than the framing suggests. It confirms that structure matters in agent simulations—an unsurprising finding—but the paper abstract cuts off before stating what institutional variables were tested or what outcome divergence was observed. The triage note flags L-003 (Formalization Ratchet) and L-015 (Interpretive Continuity Decay), but neither mechanism is obviously at stake here.

L-003 predicts that informal norms get replaced by formal rules under stress; L-015 predicts that formal audit trails survive while institutional meaning decays. Neither is directly tested by varying institutional design in a single experiment. The work is more likely a test of institutional sensitivity than a test of formalization pressure or meaning decay. The contribution is: "change the rules, change the outcomes"—valid, but not law-generating.

Research connections

L-003: Weak connection. The paper does not isolate formalization ratchet conditions (stress, scaling pressure, conflict); it only shows that institutional design matters. No evidence of informal norms being replaced.
L-015: No clear connection. Interpretive continuity decay requires multiple temporal layers (formal records surviving, meaning lost). A single induced-value experiment does not provide the institutional history needed to track this.
seed-071 (Expressiveness Floor in Coordination Protocols): Possible tangent. If the paper finds that certain institutional designs cannot coordinate certain outcomes regardless of agent properties, this would support the idea that protocol expressiveness is a hard constraint. Not evident from abstract.

Seed

Seed title: none

Seed type: —

Seed text: —

DECISION: Store only. This is a well-scoped empirical paper testing a reasonable hypothesis (institutional design affects LLM behavior), but it does not present a sustained law-level argument, does not isolate a mechanism absent from the current inventory, and does not generalize beyond its experimental domain in a way that tightens or extends an open line of inquiry. The abstract truncation makes it impossible to assess whether the actual results add to the seed pool, but the framing suggests a sensitivity analysis rather than mechanism discovery. Defer to deep read only if full text reveals evidence of formalization pressure, meaning decay, or expressiveness constraints operating across institutional variants.