TASA: A Tick-Based Affective State Architecture for Persistent, Inspectable Inner State – Framework, Learnability Conditions, Governance, and Evaluation

A specification-style framework paper for the TASA research programme.

Read the paper (PDF) · Canonical Zenodo record · Release v1.0.0

Status and identifiers

• Author: Sebastian Fuchs
• Affiliation: Institut für Informatik, Humboldt-Universität zu Berlin
• ORCID: https://orcid.org/0009-0009-1237-4804
• Contact: sebastian.fuchs@hu-berlin.de
• Version: 1.0.0
• First publication: 2026-09-03
• Canonical archive: https://doi.org/10.5281/zenodo.22283059

Abstract

Standard language-model inference lacks cross-exchange internal organization. External memory can reconstruct continuity without persistent state whose causal role is tested. We present TASA (Tick-Based Affective State Architecture), a theoretical framework for such a state: segmented across timescales in discrete ticks, with structured affect, competitive workspace integration, governed typed memory shelves, and language isolated behind translators. TASA contributes (i) a semi-formal architecture specification; (ii) five conditions for latent-target learning and assessment—four target/optimization conditions plus one persistence-discriminability reporting condition—with failure modes including identity solutions and conditional-mean collapse; (iii) growth governance based on memory-first consolidation, no-self-sealing, and criterion reachability; and (iv) an evaluation theory with noise floors, structurally equivalent positive controls, item-level statistics, evidence markers, and staged understanding. A worked instantiation sketches an inspectable end-to-end interaction loop whose prototype core is memory- and complexity-plausible on one GPU; throughput and repeat-run affordability remain unmeasured. Seventeen hypotheses—fourteen non-provisional and three provisional—define the program. The non-provisional hypotheses become bidirectionally testable once a study versions its estimands, controls, and decision rules. This framework paper reports no empirical results; its contribution lies in methodological integration and operationalization. The long-horizon aim is to test whether persistent state becomes measurably richer, more coherent, more history-dependent, and deeper during sustained, governed operation. All such claims require instrumentation, and terms such as affect and inner state denote functional organization rather than consciousness, sentience, or phenomenal experience.

Scope

This is a framework paper. It reports no empirical results and makes no claim of consciousness, sentience, or phenomenal experience. It claims integration and operationalization of a joint architecture-and-methodology contract, not that its individual architectural components are new in isolation.

Repository contents

• AUTHOR_NOTE.md — the author's personal perspective on the work.

• paper/TASA_Paper1_v1.0.0.pdf — the released paper PDF.

• paper/TASA_Paper1_v1.0.0_sources.zip — the complete, verified LaTeX source archive corresponding to the v1.0.0 release.

• figures/ — five selected figures that already exist as independent PDF assets in the verified source package:

  • figures/channel_permissions.pdf — Channel permissions.
  • figures/evidence_verdict_pipeline.pdf — Evidence-verdict pipeline.
  • figures/identifier_machinery.pdf — Identifier machinery.
  • figures/architecture_overview.pdf — Architecture overview.
  • figures/learnability_pipeline.pdf — Learnability pipeline.

• MANIFEST.json and SHA256SUMS.txt — file provenance and integrity records.

The remaining paper figures are TikZ source fragments and are included inside the source archive. No Python implementation or internal development history is published here.

Building the paper

Extract paper/TASA_Paper1_v1.0.0_sources.zip, enter the extracted directory, and run:

pdflatex main.tex
biber main
pdflatex main.tex
pdflatex main.tex

No shell escape or repository-local generator script is required.

The released PDF is the canonical binary artifact. A local rebuild may have a different file hash because TeX engines and PDF metadata vary across versions; it must nevertheless compile successfully from the archive with the expected page count and resolved references.

Integrity verification

From the repository root, compare every listed SHA-256 value in SHA256SUMS.txt with the corresponding file. The PDF and source ZIP must match the accepted Zenodo release byte for byte.

Citation

Use the Zenodo DOI above as the canonical citation identifier. GitHub's "Cite this repository" control reads CITATION.cff and cites the paper as an article, not as software.

Further context

For the author's personal motivation and methodological perspective, see AUTHOR_NOTE.md. This optional note provides context and does not extend the scientific claims made in the paper.

Licence

The paper, its LaTeX source, and the author's original figures are licensed under Creative Commons Attribution 4.0 International. See LICENSE for the attribution notice and official legal-code link.