Fooling Yourself: how narratives shape beliefs

Source: econ.GN updates on arXiv.org — https://arxiv.org/abs/2607.04753 Date read: 2026-09-02 Connected to: L-004, L-013, seed-030 Kind: empirical content Escalation: store-only Escalation rationale:

What this is

A laboratory experiment in behavioral economics testing whether nondiagnostic narrative details (statistically uninformative framing) distort Bayesian belief updating in a controlled urn-ball task. The work isolates narrative structure from statistical content to measure belief distortion under incentivized reporting conditions.

What I took from it

The paper documents a narrative capture effect — that formally equivalent information presented through different narrative frames produces divergent belief reports, even under monetary incentives for accuracy. This confirms L-004 (Goodhart Generalization) in a narrow domain: when beliefs themselves become the measured proxy for accurate state inference, the narrative wrapping of statistical evidence becomes an optimization target.

However, the scope is constrained. The mechanism appears to be individual-level cognitive bias (narrative coherence preference, attention capture) rather than a protocol-level phenomenon. The task is deliberately stripped of the coordination, verification, and enforcement machinery that characterize the "new nature" systems under study. The effect shows no evidence of propagation through multi-agent systems, ossification under repeated application, or interaction with formalization pressure — the conditions that would make this a law candidate for the funnel rather than a psychology result.

The connection to L-013 (Paradigm-Locked Anomaly Tolerance) is speculative: the paper does not examine whether organizations or protocols sustain belief distortion despite accumulating evidence of misalignment.

Research connections

L-004: Confirms metric capture in the domain of belief reporting — the proxy (reported belief) diverges from the target (accurate state inference) when narrative framing becomes legible and optimizable. But the mechanism is cognitive, not systemic.
L-013: Tangential. Paper shows individual anomaly acceptance (false belief persistence) but does not examine institutional or protocol-level tolerance for conflicting evidence.
seed-030: Narrative-driven belief formation as a coordination substrate — but the paper shows no evidence of multi-agent lock-in or protocol hardening around narrative anchors.

Seed

Seed title: Narrative Proxy Capture in Belief-Report Protocols Seed type: observation Seed text: When belief states are formalized as legible, measurable, and incentivized outputs in a protocol (e.g., prediction markets, organizational forecasting, certification systems), nondiagnostic narrative framing embedded in the information input produces measurable divergence between reported belief and diagnostic evidence. The effect persists even under direct financial incentive for accuracy, suggesting that narrative coherence or salience operates as an orthogonal optimization axis. This may generalize to any protocol that treats belief reports as decision inputs and allows narrative structure to co-evolve with the formal information signal.