Skip to content

Active Inference Battleship v0.4.0

Choose a tag to compare

@docxology docxology released this 26 Oct 16:17
· 11 commits to main since this release

Public release from the Active Inference Institute.

Play the dashboard · Read the manuscript · Zenodo archive / DOI 10.5281/zenodo.23670188

The PDF cover, abstract, repository citation file, and Zenodo metadata use the same manuscript DOI. The archive links back to this tagged release. The manuscript deposit includes its companion editions, software bundles, and reproducibility source; no separate software DOI or npm registry release is claimed.

This release includes a playable browser dashboard, diagnostics and decision explorer for either actor, real Generalized Notation Notation parser/adapter provenance, a typed core package and CLI, and a three-edition manuscript with 34 references. GNN means Generalized Notation Notation, not graph neural network.

The five-profile study separates completion shots, early hits, and label uncertainty over 256 paired units / 1,280 searches. Figures and captions distinguish Holm-adjusted sign tests from descriptive paired mean intervals. The finite generated catalog, truthful game feedback, subjective sensor noise, and one-shot objective remain explicit limitations. Publication does not establish unrestricted optimality or educational efficacy.

Downloads:

  • active_inference_battleship_combined.pdf: DOI-bearing manuscript.
  • active_inference_battleship.docx: editable Word edition.
  • active-inference-battleship-reader-v0.4.0.zip: complete HTML reader, linked research materials, and PDF/Word alternatives.
  • active-inference-battleship-dashboard-v0.4.0.zip: static browser application. Extract and serve with a static HTTP server.
  • active-inference-battleship-0.4.0.tgz: typed ESM package and headless CLI for Node/Bun.
  • active-inference-battleship-source-v0.4.0.zip: v0.4.0 code and research with a citation-only follow-up correction. The original code tag is unchanged.
  • CITATION.cff: schema-validated software metadata with the preferred manuscript citation.
  • SHA256SUMS: downloadable-file checksums.
  • RELEASE_VALIDATION.json: version/source bindings and bounded builder verification.

Fresh verification: 97 core/package tests, 47 manuscript tests, 70 browser tests, and 344 reader captures with no page errors; actual pinned GNN parsing, typecheck, build, installed Node/Bun consumers, and all-edition structure checks passed. Lint has nine retained informational advisories and no errors. Prior independent scientific replay evidence is retained through unchanged study inputs; it was not rerun for this release. LibreOffice projection is not a native Microsoft Word guarantee.

Software: MIT. Manuscript, original figures, and research data: CC BY 4.0. Dependencies and cited works retain their own rights.