Overview
Statistical power is an investigator-facing operating characteristic of an adaptive-study design. Before simulation, the investigator fixes an agent-side model, evaluator-side process, testing setting, policy, and replication plan. Each embedded agent acts only on visible history; simulated hidden truth is retained for scoring. active_inference_power makes that conditional estimand inspectable. The suite combines fixed-horizon procedures, analytic references, dependence/calibration experiments, and a discrete-state active-inference agent checked against a binary oracle. It extends this to action loops with sensing reliability, latent context, target choice, cost, and stopping. The study distinguishes model-relative posterior belief from calibrated p-value and likelihood-ratio e-process evidence. It compares Benjamini–Hochberg (BH) false discovery rate (FDR) procedures with family-wise error rate (FWER) alternatives, and separates either evidence object from online FDR procedure-specific accounting. Results are scenario-indexed finite-simulation estimates with Monte Carlo standard error (MCSE) and declared error, dependence, filtration, and optional-stopping boundaries; they do not assign a universal power value to an agent, task environment, or active inference. Instead, they support auditable comparisons among explicitly declared adaptive-study designs. Contracts, seed schedules, certificates, figures, claim ledger, and rendered manuscript form a linked evidence chain, allowing readers to trace each claim to its design and artifact. Source and release materials are available at the verified GitHub repository ActiveInferenceInstitute/active_inference_power.
Active Inference Power Suite v1.0.0 is a source-bound release of an adaptive-study design suite. It reports scenario- and policy-indexed finite-simulation operating characteristics; it does not claim a universal "power of active inference."
Source repository and exact release: https://github.com/ActiveInferenceInstitute/active_inference_power/releases/tag/v1.0.0
Concept DOI (release family): 10.5281/zenodo.21695160 (https://doi.org/10.5281/zenodo.21695160)
Version DOI (this immutable archive): 10.5281/zenodo.21695161 (https://doi.org/10.5281/zenodo.21695161)
Zenodo record: https://zenodo.org/records/21695161
PDF SHA-256: 24fa25a4f29affcfd92c8c001ff6487a0c36960c6b4f38ed4419984fc8743cbf
The uploaded PDF, exact tag-derived source archive, release manifest, renderer provenance, and final review receipt make the release auditable. The evidence boundary remains finite, scenario-specific simulation rather than a universal theorem or deployment claim.
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- In the configured design, FWER-oriented procedures reached power 0.315–0.321 while FDR-oriented procedures reached 0.430–0.749, framed as the expected trade-off, not dominance.
- Storey's paired power gain over BH was about 0.073, which the paper stresses is an operating-characteristic comparison, not a validity certificate.
- Under weak evidence the posterior-threshold policy had FDR 0.153 and FWER 0.366, while the calibrated BH policy made no rejections, illustrating a calibration gap.
- Across the dependence grid, BH estimates stayed within the finite-simulation band and BY stayed more conservative; the paper states this is not a new PRDS proof.
- The paper concludes power numbers should not be carried into a new design; adaptive actions change data path, cost and evidence contract together.
Methods
- Power estimated as an operating characteristic of an investigator-declared design — Defines power over replications of a declared model/process/setting/policy design, scoring traces against evaluator-only hidden truth.
- Seeded Monte Carlo comparison of 9 FWER/FDR correction procedures — Compares Bonferroni, Šidák, Holm, Hochberg, BH, BY, Storey, adaptive BH and weighted BH on a seeded two-groups design with MC SE bands.
- Dependence-regime stress grid for BH and BY — Evaluates BH and BY FDR and power under negative equicorrelation, independence, positive-factor and block covariance regimes.
- Discrete-state active-inference agent: posterior threshold vs BH on calibrated p-values — Runs identical streams through a posterior-threshold decision and a BH-calibrated evidence rule under strong and weak evidence regimes.
- Action-in-the-loop policies with sensing reliability, cost and stopping — Compares fixed-horizon, posterior-cutoff, information-gain, cost-aware, posterior-sampling and e-process stopping policies in a synthetic action loop.
Summary sources: Paper metadata and evidence · Extracted source text.
Document downloads
PDFs are hosted here; supporting sources may use pinned archives.
- active_inference_power_v1.0.0.pdf PDF · 10.69 MiB · extracted-text source
Citation
Citation metadata follows the unified bibliography.
Catalog details and resources
Related works
Ranked by recorded methods, topics, contributors and associated software, with contributor roles and domain-only matches labelled. Metadata overlap is not a citation or a scientific equivalence.