Prediction-error tolerance: Error attribution as the hinge between inquiry and the self
Abstract
Close to one in three early-career researchers reports elevated psychological distress, at rates varying little across disciplines, genders and career stages. The standard explanation attributes the burden to the conditions of academic work; its rival prescribes resilience. Neither says what a condition does to a researcher who cannot yet tell whether an unexpected result reflects the phenomenon, the instrument or her own competence. A prediction error has no address until a researcher gives it one: by the Duhem–Quine thesis it may belong to the object of inquiry, the method or the context of evaluation, and logic does not decide which. We define prediction-error tolerance as the capacity to keep treating discrepancies as operable questions while holding uncertainty open, and identify its core operation, error-attribution separation, as the point at which two costs are decided together. When an error is routed to the self instead of to its address, the researcher’s well-being is damaged and the error is lost to inquiry; the adaptations that follow (safe questions, discarded anomalies, work subdivided to fit the measure) are those metascience explains by incentives alone. The two costs, we propose, are effects of one regime with a shared mediating step. We distinguish the construct from ten neighbours, fix the error episode as the unit of measurement, and state six predictions with the observations that would defeat each, including one that would leave the incentive account standing and ours not. An environment that returns errors with an address protects the person and the inquiry together.
Related
- Feedback Valence, Prediction Error, and the Limits of Computational Attribution in Symptom-Related EEG
- Learning from Error Predictions Rather Than Prediction Errors: A Perspective on Self-Supervised Learning in the Brain
- Cross-Frequency Coupling as a Neural Substrate for Prediction Error Evaluation: A Laminar Neural Mass Modeling Approach
- Deciphering the dynamic genetic architecture of adult-plant salinity tolerance in bread wheat using multi-model GWAS and genomic prediction.
- Supervised Domain Adaptation Mitigates Cross-Ethnicity Prediction Error in Neuroimaging Based Cognitive Prediction
BibTeX
@misc{sakai2026predictionerror,
title = {{Prediction-error tolerance: Error attribution as the hinge between inquiry and the self}},
author = {Tomoko Sakai and Kazuhiro Hayashi and Norihiro Sadato},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/j6rct_v1},
url = {https://osf.io/preprints/psyarxiv/j6rct_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Prediction-error tolerance: Error attribution as the hinge between inquiry and the self
AU - Tomoko Sakai
AU - Kazuhiro Hayashi
AU - Norihiro Sadato
PY - 2026
DA - 2026/09/27
PB - PsyArXiv
DO - 10.31234/osf.io/j6rct_v1
UR - https://osf.io/preprints/psyarxiv/j6rct_v1/
AB - Close to one in three early-career researchers reports elevated psychological distress, at rates varying little across disciplines, genders and career stages. The standard explanation attributes the burden to the conditions of academic work; its rival prescribes resilience. Neither says what a condition does to a researcher who cannot yet tell whether an unexpected result reflects the phenomenon, the instrument or her own competence. A prediction error has no address until a researcher gives it one: by the Duhem–Quine thesis it may belong to the object of inquiry, the method or the context of evaluation, and logic does not decide which. We define prediction-error tolerance as the capacity to keep treating discrepancies as operable questions while holding uncertainty open, and identify its core operation, error-attribution separation, as the point at which two costs are decided together. When an error is routed to the self instead of to its address, the researcher’s well-being is damaged and the error is lost to inquiry; the adaptations that follow (safe questions, discarded anomalies, work subdivided to fit the measure) are those metascience explains by incentives alone. The two costs, we propose, are effects of one regime with a shared mediating step. We distinguish the construct from ten neighbours, fix the error episode as the unit of measurement, and state six predictions with the observations that would defeat each, including one that would leave the incentive account standing and ours not. An environment that returns errors with an address protects the person and the inquiry together.
ER -