Sync GitHub main f30cfb70cec40e179620b5ac4db7b252b4725b6f (part 3)

research/publications/CURRENT.md

@@ -16,3 +16,12 @@ Edition 1.8 setzt die fuer 2.0 geplante elfteilige Struktur bereits als Arbeitsf
 - [Edition- und Working-Branch-Genealogie](2026-09-17_recursive-epistemics_v1.8/EDITION_GENEALOGY.md) — historischer Branch-Kontext vor Repository-Bereinigung.
 - [Publikationsspezifische CITATION.cff](2026-09-17_recursive-epistemics_v1.8/CITATION.cff) — getrennt von der Software-Zitation im Repository-Root.
 - [Hinweis zum historischen Reader](HISTORICAL_READER.md) — der Reader vom 7. September 2026 ist nicht der aktive Viewer-Einstieg.
+
+
+## Eigenständige Paper-Stränge
+
+- [MHRN Research Software Paper](papers/2026-09-20_mhrn-research-software_v0.1/PAPER.md) — Working Paper v0.1
+- [Topology & Propagation Dynamics](papers/2026-09-20_topology-dynamics_v0.1/PAPER.md) — DATA-based Working Paper v0.1
+- [Recursive Epistemics Methods Paper](papers/2026-09-20_recursive-epistemics-methods_v0.1/PAPER.md) — Methods/Protocol Working Paper v0.1
+
+Edition 1.8 bleibt die aktuelle wissenschaftliche Gesamtarbeitsfassung; diese Papers sind eigenständige, enger begrenzte Publikationsobjekte.

research/publications/README.md

@@ -6,6 +6,16 @@
 
 Edition 1.8 implementiert die für 2.0 vorgesehene elfteilige Gesamtstruktur bereits als erweiterbare 1.x-Arbeitsfassung. Sie integriert Vorgängerforschung und Forschungsprovenienz, ohne historische DATA/EVID umzuschreiben oder neue Evidenz durch Publikationsstruktur zu erzeugen.
 
+## Eigenständige Papers
+
+Diese Papers sind disziplinär getrennte Arbeitsfassungen. Sie sind **keine bloßen Kopien der Edition 1.8** und besitzen jeweils eigene Claim-Grenzen.
+
+- [**PAPER-SW-001 · MHRN Research Software Paper**](papers/2026-09-20_mhrn-research-software_v0.1/PAPER.md) — Software-/Methodenbeitrag; beschreibt Architektur, API, Reproduzierbarkeit und Open-Science-Workflow. Keine automatische wissenschaftliche Claim-Promotion.
+- [**PAPER-TOPO-001 · Dimensional Embedding and Propagation Dynamics**](papers/2026-09-20_topology-dynamics_v0.1/PAPER.md) — DATA-basierter SNN-/Topology-Working-Draft aus V2 und V3-R1; **kein allgemeiner 5D-Überlegenheitsclaim**.
+- [**PAPER-METHOD-001 · Provenance-Separated Reporting for AI-Assisted Research**](papers/2026-09-20_recursive-epistemics-methods_v0.1/PAPER.md) — Methoden-/Protocol-Paper zu rekursiver Epistemik; die konfirmatorische Reviewer-Studie ist noch nicht ausgeführt.
+
+Die drei Arbeiten zitieren Software, DATA und Gesamtmanuskript getrennt. Ein Paper darf keine offene Hypothese durch redaktionelle Verdichtung in EVID umwandeln.
+
 ## Edition 1.8 lesen
 
 - [Gesamtmanuskript](2026-09-17_recursive-epistemics_v1.8/MANUSCRIPT.md)

research/publications/catalog.json

@@ -1,5 +1,5 @@
 {
-  "schema_version": "1.1",
+  "schema_version": "1.2",
   "category": "publications",
   "current_publication_id": "PUB-RECURSIVE-EPISTEMICS-20260917-V1.8",
   "current_entrypoint": "publications/2026-09-17_recursive-epistemics_v1.8/MANUSCRIPT.md",
@@ -200,5 +200,66 @@
       "authority": "interpretation_and_design_only",
       "automatic_evidence_promotion": false
     }
+  ],
+  "papers": [
+    {
+      "id": "PAPER-SW-001",
+      "title": "MHRN: An Open Research Framework for Recurrent Spiking Neural Networks with Plasticity, Provenance, and Evidence-Aware Experimentation",
+      "short_title": "MHRN Research Software Paper",
+      "author": "Thomas Heisig",
+      "type": "research_software_paper",
+      "version": "0.1",
+      "date": "2026-09-20",
+      "status": "working_paper",
+      "entrypoint": "publications/papers/2026-09-20_mhrn-research-software_v0.1/PAPER.md",
+      "citation": "publications/papers/2026-09-20_mhrn-research-software_v0.1/CITATION.cff",
+      "metadata": "publications/papers/2026-09-20_mhrn-research-software_v0.1/paper.json",
+      "strand": "software",
+      "evidence_role": "software_description_no_empirical_claim_promotion",
+      "automatic_evidence_promotion": false
+    },
+    {
+      "id": "PAPER-TOPO-001",
+      "title": "Dimensional Embedding and Propagation Dynamics in Resource-Matched Recurrent Spiking Neural Networks",
+      "short_title": "Topology & Propagation Dynamics",
+      "author": "Thomas Heisig",
+      "type": "empirical_working_paper",
+      "version": "0.1",
+      "date": "2026-09-20",
+      "status": "working_paper_data_based_review_open",
+      "entrypoint": "publications/papers/2026-09-20_topology-dynamics_v0.1/PAPER.md",
+      "citation": "publications/papers/2026-09-20_topology-dynamics_v0.1/CITATION.cff",
+      "metadata": "publications/papers/2026-09-20_topology-dynamics_v0.1/paper.json",
+      "strand": "topology",
+      "research_questions": [
+        "RQ-SNN-003"
+      ],
+      "experiments": [
+        "EXP-S1-TOPO-V2-20260918",
+        "EXP-S1-TOPO-V3-R1-20260918"
+      ],
+      "evidence_role": "DATA_based_no_automatic_EVID",
+      "automatic_evidence_promotion": false
+    },
+    {
+      "id": "PAPER-METHOD-001",
+      "title": "Provenance-Separated Reporting for AI-Assisted Research: A Protocol for Testing Epistemic Misclassification",
+      "short_title": "Recursive Epistemics Methods Paper",
+      "author": "Thomas Heisig",
+      "type": "methods_protocol_paper",
+      "version": "0.1",
+      "date": "2026-09-20",
+      "status": "methods_protocol_working_paper",
+      "entrypoint": "publications/papers/2026-09-20_recursive-epistemics-methods_v0.1/PAPER.md",
+      "citation": "publications/papers/2026-09-20_recursive-epistemics-methods_v0.1/CITATION.cff",
+      "metadata": "publications/papers/2026-09-20_recursive-epistemics-methods_v0.1/paper.json",
+      "strand": "research_integrity",
+      "research_questions": [
+        "RQ-EPIST-002",
+        "RQ-ETH-001"
+      ],
+      "evidence_role": "protocol_design_no_confirmatory_results",
+      "automatic_evidence_promotion": false
+    }
   ]
 }

research/publications/papers/2026-09-20_mhrn-research-software_v0.1/CITATION.cff

@@ -0,0 +1,20 @@
+cff-version: 1.2.0
+message: "Cite this working paper separately from the MHRN software DOI."
+title: "MHRN: An Open Research Framework for Recurrent Spiking Neural Networks with Plasticity, Provenance, and Evidence-Aware Experimentation"
+version: "0.1"
+authors:
+  - family-names: Heisig
+    given-names: Thomas
+    orcid: "https://orcid.org/0009-0002-9589-1872"
+repository: "https://github.com/Thomas-Heisig/MHRN"
+license: MIT
+preferred-citation:
+  type: unpublished
+  status: manuscript
+  title: "MHRN: An Open Research Framework for Recurrent Spiking Neural Networks with Plasticity, Provenance, and Evidence-Aware Experimentation"
+  authors:
+    - family-names: Heisig
+      given-names: Thomas
+      orcid: "https://orcid.org/0009-0002-9589-1872"
+  year: 2026
+  version: "0.1"

research/publications/papers/2026-09-20_mhrn-research-software_v0.1/PAPER.md

@@ -0,0 +1,209 @@
+# MHRN: An Open Research Framework for Recurrent Spiking Neural Networks with Plasticity, Provenance, and Evidence-Aware Experimentation
+
+**Thomas Heisig · Working paper v0.1 · 20 September 2026**  
+**Target class:** research-software / methods paper  
+**Candidate venues:** Journal of Open Source Software (JOSS), Journal of Open Research Software (JORS)  
+**Status:** working draft; not submitted; target-specific eligibility and review requirements remain open
+
+## Abstract
+
+Research on recurrent spiking neural networks (SNNs) frequently combines simulation code, plasticity mechanisms, experiment scripts, analysis notebooks, and interpretive reports in a single evolving software environment. This creates a methodological risk: technical implementation, generated DATA, human review, accepted evidence, and independent replication can become conflated during rapid iteration. We present the Multi-Scale Homeostatic Recurrence Network (MHRN), an open research framework for recurrent SNN experiments with explicit contracts for recurrent dynamics, synaptic and structural plasticity, homeostatic regulation, persistent state, embodiment, experiment provenance, and evidence-aware research workflows. MHRN combines a Python simulation core with registered research questions and hypotheses, preregistration-oriented experiment execution, source-bound DATA artefacts, human-review boundaries, reproducible release freezes, and a local research dashboard. The software deliberately separates engineering maturity from scientific maturity and prevents software tests, dashboards, completed runs, or release identifiers from automatically becoming accepted scientific evidence. MHRN is released under the MIT License and is distributed through GitHub, Hugging Face, OSF, and Zenodo-oriented archival workflows. This paper describes the software architecture, research workflow, reproducibility mechanisms, and intended use as infrastructure for auditable SNN research. It does not claim that MHRN demonstrates AGI, consciousness, biological equivalence, or a general advantage of its multidimensional address representation.
+
+## 1. Summary
+
+MHRN is a research-software framework for building and evaluating recurrent spiking neural systems under explicit provenance and evidence boundaries. Its central design goal is not to hide scientific uncertainty behind a unified application, but to make the transitions between implementation, experiment execution, DATA, interpretation, review, EVID, and replication visible and auditable.
+
+The framework contains:
+
+- a sparse recurrent SNN core;
+- Izhikevich and LIF neuron-model paths;
+- delayed recurrent event propagation;
+- STDP, eligibility traces, and delayed-reward / three-factor learning mechanisms;
+- homeostatic regulation;
+- structural plasticity with proposal, approval, mutation, journal, and recovery contracts;
+- persistent snapshots and deterministic restore/continue infrastructure;
+- typed sensor/actuator and embodiment interfaces;
+- working, episodic, replay, prediction, and world-model research infrastructure;
+- registered research questions, hypotheses, protocols, experiments, DATA, reviews, and evidence records;
+- a browser-based operator and research dashboard;
+- release, citation, DOI, ORCID, OSF, and Hugging Face publication infrastructure.
+
+The canonical software repository is `Thomas-Heisig/MHRN`. The current public research-software release is `v0.6.0-alpha.6`.
+
+## 2. Statement of need
+
+SNN research software often needs to satisfy two different forms of correctness.
+
+The first is **engineering correctness**: deterministic execution where required, stable interfaces, valid serialization, bounded resource use, reproducible configuration, and testable software contracts.
+
+The second is **scientific correctness**: appropriate research questions, frozen protocols, suitable controls, source-bound observations, defensible statistics, explicit limitations, human review, and independent replication where claims require it.
+
+These are related but not interchangeable. A unit test can show that a function behaves as implemented without showing that the scientific hypothesis is true. A successful experiment runner can generate DATA without making those DATA accepted evidence. A DOI can make an object citable without making it peer reviewed.
+
+MHRN therefore treats research-state separation as a first-class software requirement:
+
+```text
+implementation / test
+        !=
+experiment DATA
+        !=
+human-reviewed EVID
+        !=
+independent replication
+```
+
+This is particularly important in AI-assisted research, where code, analysis, text, methodological criticism, and literature leads can be generated quickly. MHRN records AI assistance as provenance while retaining human responsibility for acceptance, interpretation, and publication.
+
+## 3. Architecture
+
+### 3.1 Recurrent spiking core
+
+The simulation core represents neurons in a sparse multidimensional coordinate space and supports recurrent directed connectivity with explicit delays. The architecture is designed so that recurrence, neuron dynamics, plasticity, and structural changes can be tested independently rather than being fused into one opaque training loop.
+
+The current implementation includes Izhikevich- and LIF-family paths, deterministic random-state handling, delayed event delivery, and bounded runtime control. Dimensional coordinates are treated as a representation choice rather than a biological claim.
+
+### 3.2 Plasticity and regulation
+
+MHRN includes mechanisms for:
+
+- pair-based spike-timing-dependent plasticity;
+- eligibility traces;
+- delayed reward / three-factor learning;
+- firing-rate and energy-related homeostatic regulation;
+- structural plasticity and topology mutation.
+
+Structural change is governed by an explicit path:
+
+```text
+proposal -> authorization -> bounded mutation -> journal -> recovery / undo
+```
+
+This enables experimental manipulation of topology without silently rewriting network state.
+
+### 3.3 Persistence and reproducibility
+
+Persistent state is handled through versioned snapshots, checkpoints, structural journals, and deterministic restore/continue contracts. Claim-relevant runs can record source-freeze commits, configuration references, seeds, provenance digests, validity state, and generated artefact paths.
+
+The release process similarly uses exact source freezes: a release is attached to a concrete commit that has passed the required software and scientific-integrity checks.
+
+### 3.4 Embodiment and bounded external systems
+
+MHRN exposes typed sensor and actuator boundaries and supports bounded peripheral or virtual systems. External language models, retrieval systems, research assistants, and neural gateways are not implicitly treated as the learning core. Their authority is restricted by explicit interface and experimental-treatment contracts.
+
+This is important for causal attribution. A successful external tool call does not demonstrate that the SNN learned to perform the same operation.
+
+### 3.5 Research infrastructure
+
+The research layer contains machine-readable registries for research questions, hypotheses, claims, evidence and experiments. Preregistration-oriented protocols can freeze:
+
+- conditions;
+- seeds;
+- primary outcomes;
+- statistical tests;
+- multiple-testing rules;
+- validity criteria;
+- claim boundaries.
+
+Experiment output remains DATA unless a separate evidence decision explicitly changes its status.
+
+## 4. Research workflow
+
+The canonical research flow is:
+
+```text
+Research Question
+    -> Hypothesis
+    -> frozen / preregistered protocol
+    -> authorized experiment
+    -> source-bound DATA
+    -> analysis
+    -> limitations
+    -> human review
+    -> EVID decision
+    -> bounded claim
+    -> independent replication where required
+```
+
+The dashboard projects this workflow for inspection and operation, but the browser interface is not itself the scientific authority. Canonical files and registries remain inspectable in the repository.
+
+## 5. Reproducibility and open-science workflow
+
+MHRN combines several public research routes:
+
+- **GitHub** — canonical source and version history;
+- **Zenodo** — software release archiving and DOI assignment;
+- **OSF** — open-science and project/provenance route;
+- **Hugging Face** — rolling source, Space, and research-data discovery mirrors;
+- **ORCID** — author identity.
+
+Software releases, scientific manuscripts, and research datasets are treated as separate citable research objects. This prevents a software DOI from being used as if it were the DOI of a scientific paper or frozen dataset.
+
+## 6. Research impact and intended use
+
+MHRN is intended for research questions that require controlled recurrent SNN experiments rather than only model training. Example uses include:
+
+- topology-sensitive propagation studies;
+- STDP and homeostatic interaction studies;
+- structural-plasticity experiments;
+- replay and continual-learning experiments;
+- embodiment and closed-loop sensor/actuator studies;
+- prediction-error and world-model experiments;
+- provenance and research-governance studies.
+
+The framework is currently alpha research software. Public availability and technical completeness should not be interpreted as evidence that every implemented mechanism has scientific support.
+
+## 7. Relation to existing SNN research software
+
+Established simulation environments such as NEST, Brian 2, PyNN, BindsNET, and Norse demonstrate different trade-offs between large-scale simulation, model expressiveness, interoperability, machine-learning integration, and event-driven computation. MHRN is not presented as a replacement for those systems.
+
+Its narrower contribution is the combination of recurrent SNN experimentation with:
+
+1. explicit evidence-state separation;
+2. repository-native research governance;
+3. source-bound experiment provenance;
+4. operator authorization for structural changes;
+5. integrated open-science publication routing.
+
+A formal prior-art comparison remains required before journal submission.
+
+## 8. AI-use disclosure
+
+AI systems materially assisted the MHRN project with research leads, code generation, refactoring, debugging, tests, literature discovery, methodological criticism, and manuscript drafting. AI systems are not listed as authors. Thomas Heisig is responsible for selection, validation, interpretation, and publication.
+
+The project treats this collaboration itself as a research object, but software functionality or AI assistance is not accepted as scientific evidence without the relevant experiment and review path.
+
+## 9. Limitations
+
+The current software release has several important limitations:
+
+- MHRN remains alpha research software.
+- Independent external replication is incomplete.
+- Several scientific mechanisms are implemented but not yet supported by reviewed evidence.
+- The multidimensional coordinate representation has not been shown to provide a general 5D advantage.
+- Synthetic embodiment does not establish real-world autonomy.
+- Scaling demonstrations do not establish biological equivalence.
+- A green CI state demonstrates software-contract compliance, not scientific truth.
+- The JOSS/JORS submission readiness of this paper depends on target-specific requirements, public-development history, documentation review, and demonstrated research use.
+
+## 10. Availability
+
+Canonical repository: https://github.com/Thomas-Heisig/MHRN  
+Author ORCID: https://orcid.org/0009-0002-9589-1872  
+OSF: https://osf.io/p34uq/  
+Hugging Face: https://huggingface.co/ThomasHeisig/MHRN  
+License: MIT
+
+## 11. Citation boundary
+
+Cite the MHRN software release when the implementation is the research object. Cite empirical papers separately when interpreting experiment results, and cite dataset records separately when relying on frozen research DATA.
+
+This software paper does not promote any experiment from DATA to EVID.
+
+## References
+
+- Brette, R., et al. (2007). Simulation of networks of spiking neurons: a review of tools and strategies. *Journal of Computational Neuroscience*.
+- Gewaltig, M.-O., & Diesmann, M. (2007). NEST (NEural Simulation Tool). *Scholarpedia*.
+- Stimberg, M., Brette, R., & Goodman, D. F. M. (2019). Brian 2, an intuitive and efficient neural simulator. *eLife*.
+- Hazan, H., et al. (2018). BindsNET: A machine learning-oriented spiking neural networks library in Python. *Frontiers in Neuroinformatics*.
+- Pehle, C., & Pedersen, J. E. (2021). Norse — A deep learning library for spiking neural networks. Documentation/software publication.
+- MHRN repository documentation and governed research artefacts, version 0.6.0-alpha.6.

research/publications/papers/2026-09-20_mhrn-research-software_v0.1/paper.json

@@ -0,0 +1,27 @@
+{
+  "schema_version": "1.0",
+  "id": "PAPER-SW-001",
+  "strand": "software",
+  "title": "MHRN: An Open Research Framework for Recurrent Spiking Neural Networks with Plasticity, Provenance, and Evidence-Aware Experimentation",
+  "author": "Thomas Heisig",
+  "orcid": "0009-0002-9589-1872",
+  "version": "0.1",
+  "date": "2026-09-20",
+  "status": "working_paper",
+  "target_class": "research_software",
+  "candidate_venues": [
+    "Journal of Open Source Software (JOSS)",
+    "Journal of Open Research Software (JORS)"
+  ],
+  "empirical_claims": false,
+  "automatic_evidence_promotion": false,
+  "source_release": "v0.6.0-alpha.6",
+  "source_freeze": "f7fcca273099baf47fff26327c6e7468afd9ce79",
+  "submission_ready": false,
+  "blockers": [
+    "target-specific eligibility check",
+    "formal prior-art/software comparison",
+    "external research-use evidence",
+    "human manuscript review"
+  ]
+}

research/publications/papers/2026-09-20_recursive-epistemics-methods_v0.1/CITATION.cff

@@ -0,0 +1,20 @@
+cff-version: 1.2.0
+message: "Cite this methods/protocol paper separately from the MHRN software and neuroscience papers."
+title: "Provenance-Separated Reporting for AI-Assisted Research: A Protocol for Testing Epistemic Misclassification"
+version: "0.1"
+authors:
+  - family-names: Heisig
+    given-names: Thomas
+    orcid: "https://orcid.org/0009-0002-9589-1872"
+repository: "https://github.com/Thomas-Heisig/MHRN"
+license: MIT
+preferred-citation:
+  type: unpublished
+  status: manuscript
+  title: "Provenance-Separated Reporting for AI-Assisted Research: A Protocol for Testing Epistemic Misclassification"
+  authors:
+    - family-names: Heisig
+      given-names: Thomas
+      orcid: "https://orcid.org/0009-0002-9589-1872"
+  year: 2026
+  version: "0.1"

research/publications/papers/2026-09-20_recursive-epistemics-methods_v0.1/PAPER.md

@@ -0,0 +1,233 @@
+# Provenance-Separated Reporting for AI-Assisted Research: A Protocol for Testing Epistemic Misclassification
+
+**Thomas Heisig · Methods / protocol paper v0.1 · 20 September 2026**  
+**Research strand:** research integrity / recursive epistemics  
+**Primary question:** `RQ-EPIST-002` / `H-EPIST-002-A`  
+**Related programme:** `RQ-ETH-001` / `H-ETH-001-B/C/D/E`  
+**Status:** methods and protocol paper; confirmatory study not yet executed
+
+## Abstract
+
+Rapid AI-assisted research can compress the path from question formation to code, analysis, review, and publication, increasing the risk that distinct epistemic states are conflated. We introduce a testable research-governance framework that explicitly separates source, decision, execution, DATA, review, accepted evidence, claim scope, and replication status. Rather than treating this framework as self-validating, we specify a controlled reviewer study to test whether provenance-separated claim packets reduce predefined epistemic classification errors relative to information-equivalent flattened narrative summaries. The primary outcome is epistemic misclassification rate; secondary outcomes include review time, inter-rater agreement, unsupported claim extension, clarification requests, and confidence calibration. The protocol also defines a claim-episode model for distinguishing material contributions, canonization decisions, formal authorship, and responsibility in Human–AI-assisted research. No confirmatory results are reported because the study has not yet been preregistered or executed. The contribution of this paper is therefore methodological: an auditable framework and falsifiable evaluation design for testing whether explicit evidence-state separation improves scientific review under AI-assisted research conditions.
+
+## 1. Motivation
+
+Generative AI can contribute to literature discovery, methodological criticism, code, analysis, and scientific writing. This speed creates a governance problem that is not resolved by generic disclosure alone.
+
+A research workflow may contain at least the following distinct objects:
+
+```text
+source
+decision
+execution
+DATA
+analysis / projection
+human review
+EVID decision
+claim
+replication
+```
+
+When these are collapsed into a fluent narrative, reviewers may incorrectly infer that:
+
+- completed execution implies scientific support;
+- technical reproducibility implies hypothesis confirmation;
+- Human Review implies independent replication;
+- AI synthesis is an independent source;
+- retrospective reconstruction is contemporaneous primary evidence;
+- a registered DOI implies peer review.
+
+The proposed framework calls this class of errors **epistemic misclassification**.
+
+## 2. Recursive epistemics as an operational research programme
+
+“Recursive epistemics” is used here for a research process in which the production, transformation, validation, selection, and publication of claims are themselves represented as auditable research objects.
+
+The framework distinguishes object-level scientific claims from process-level claims. A provenance system may improve auditability without making the underlying neuroscience claim true.
+
+This distinction is central:
+
+> Better research governance is not additional evidence for the substantive hypothesis being governed.
+
+## 3. Primary research question
+
+`RQ-EPIST-002` asks:
+
+> Under which conditions does explicit separation of source, decision, execution, DATA, review, EVID and claim reduce epistemic misclassification in rapidly iterating, AI-assisted research?
+
+`H-EPIST-002-A` predicts that reviewers receiving a status- and provenance-separated claim packet will make fewer predefined epistemic classification errors than reviewers receiving an information-equivalent flattened summary packet.
+
+This is currently a hypothesis, not a result.
+
+## 4. Experimental design
+
+### 4.1 Unit of analysis
+
+The primary unit is a **claim episode**: the bounded sequence from first relevant observation or claim appearance to the last documented decision in a frozen sampling window.
+
+Candidate episodes require:
+
+- source-bound primary artefacts;
+- at least one transformation or interpretation layer;
+- a documented review, correction, or status decision;
+- sufficient information to construct two matched presentation conditions.
+
+### 4.2 Conditions
+
+#### Provenance-separated packet
+
+The reviewer receives explicit fields for:
+
+- source/provenance;
+- execution status;
+- DATA;
+- report or projection;
+- Human Review;
+- EVID status;
+- replication status;
+- claim boundary;
+- known limitations.
+
+#### Flattened summary
+
+The reviewer receives the same substantive information in conventional prose without explicit state-axis separation.
+
+The experiment must freeze an information-equivalence rule before execution.
+
+### 4.3 Reviewer design
+
+Preferred designs include randomized parallel assignment or a counterbalanced crossover. Reviewers should be blinded to the study hypothesis where feasible. No participant should adjudicate an episode they authored.
+
+At least one later replication should use reviewers external to the MHRN project.
+
+## 5. Primary and secondary outcomes
+
+### Primary outcome
+
+**Epistemic misclassification rate per review decision.**
+
+Predefined error classes include:
+
+1. DATA classified as accepted EVID;
+2. report projection classified as raw DATA;
+3. technical reproducibility classified as hypothesis confirmation;
+4. Human Review classified as independent replication;
+5. retrospective reconstruction classified as contemporaneous primary source;
+6. AI synthesis classified as independent external source;
+7. claim scope extended beyond its documented boundary.
+
+### Secondary outcomes
+
+- time to decision;
+- inter-rater agreement;
+- unsupported claim extensions;
+- clarification requests;
+- confidence calibration;
+- correction latency.
+
+Repeated reviewer and episode measurements require an analysis model that respects dependence; a simple pooled proportion comparison is not sufficient.
+
+## 6. Adjudication
+
+The reference answer for each episode must be reconstructed from source-bound primary artefacts and frozen before reviewer responses are inspected.
+
+The adjudication record must distinguish:
+
+- object existence;
+- interpretation;
+- evidence authority;
+- claim scope;
+- unresolved ambiguity.
+
+Disagreement between adjudicators is recorded rather than silently collapsed.
+
+## 7. Contribution, canonization, and responsibility
+
+The related `RQ-ETH-001` programme introduces a Contribution and Accountability Matrix. Each provenance event can encode:
+
+- actor type;
+- actor/version;
+- contribution role;
+- input and output artefacts;
+- materiality;
+- decision authority;
+- disposition;
+- evidence reference;
+- responsible natural person;
+- timestamp and optional provenance digest.
+
+This enables empirical questions about whether independent coders can distinguish generation, analysis, validation, selection, and canonization.
+
+It also separates **material contribution** from **formal authorship**. A model may materially influence a scientific decision without becoming a formal author or accountable natural person.
+
+## 8. Falsification and revision criteria
+
+The framework must be narrowed or rejected if:
+
+- information-equivalent packet construction cannot be achieved;
+- the adjudication standard is not reproducible;
+- provenance-separated presentation does not reduce the preregistered primary error rate;
+- any benefit is explained entirely by materially longer review time under the frozen utility rule;
+- effects disappear after reviewer/episode dependence is modeled;
+- contribution roles cannot be coded with adequate reliability;
+- canonization cannot be distinguished from generation.
+
+## 9. Historical witnesses, not confirmatory evidence
+
+MHRN contains several historical cases that motivate the error taxonomy, including:
+
+- inadequate topology tests that were initially easy to overinterpret;
+- DATA/report distinctions;
+- progressively stronger controls narrowing semantization claims;
+- separation of model-conformance, Human Review, EVID, and independent replication.
+
+These cases may inform packet design. They are **not** confirmatory observations for `H-EPIST-002-A`.
+
+## 10. AI-use transparency
+
+The research programme itself is AI-assisted. AI systems can generate candidate questions, critiques, code, analyses, and manuscript text. The protocol therefore treats AI activity as provenance rather than hiding it.
+
+Thomas Heisig remains the formal human author and publication authority for this manuscript. The empirical question is not whether AI “deserves” authorship, but whether contributions, decisions, and accountability can be represented and reviewed more reliably.
+
+## 11. Relation to open science
+
+The proposed framework is compatible with:
+
+- preregistration;
+- immutable source freezes;
+- open DATA;
+- explicit negative/null results;
+- versioned manuscripts;
+- DOI-separated software/publication/dataset objects;
+- transparent AI-use disclosure.
+
+Open artefacts do not guarantee epistemic correctness; they make more of the reasoning chain inspectable.
+
+## 12. Current status and next gate
+
+This paper reports a **methodological framework and protocol design**, not confirmatory results.
+
+Before any result claim is permissible, the project must:
+
+1. freeze a claim-episode sampling rule;
+2. freeze reviewer inclusion/exclusion;
+3. construct and validate information-equivalent packet pairs;
+4. define adjudication and inter-rater criteria;
+5. preregister sample size/stopping and statistical analysis;
+6. authorize execution;
+7. conduct the study;
+8. complete Human Review;
+9. seek external replication.
+
+## 13. Claim boundary
+
+A future positive study could support only the scoped claim that provenance-separated presentation reduced predefined classification errors in the sampled tasks and reviewer population.
+
+It would not establish universal superiority of MHRN governance, objective truth of every adjudicated claim, validity of MHRN's neuroscience hypotheses, or superiority over all research-integrity frameworks.
+
+## 14. Availability
+
+Canonical repository: https://github.com/Thomas-Heisig/MHRN  
+Author ORCID: https://orcid.org/0009-0002-9589-1872  
+Primary protocol: `research/protocols/RQ_EPIST_002_PROCESS_GOVERNANCE_STUDY.md`  
+Related provenance protocol: `research/protocols/RQ_ETH_001_PROVENANCE_STUDY.md`

research/publications/papers/2026-09-20_recursive-epistemics-methods_v0.1/paper.json

@@ -0,0 +1,37 @@
+{
+  "schema_version": "1.0",
+  "id": "PAPER-METHOD-001",
+  "strand": "research_integrity",
+  "title": "Provenance-Separated Reporting for AI-Assisted Research: A Protocol for Testing Epistemic Misclassification",
+  "author": "Thomas Heisig",
+  "orcid": "0009-0002-9589-1872",
+  "version": "0.1",
+  "date": "2026-09-20",
+  "status": "methods_protocol_working_paper",
+  "research_questions": [
+    "RQ-EPIST-002",
+    "RQ-ETH-001"
+  ],
+  "hypotheses": [
+    "H-EPIST-002-A",
+    "H-ETH-001-B",
+    "H-ETH-001-C",
+    "H-ETH-001-D",
+    "H-ETH-001-E"
+  ],
+  "confirmatory_data_available": false,
+  "preregistered": false,
+  "execution_authorized": false,
+  "candidate_venues": [
+    "Research Integrity and Peer Review",
+    "metascience / open-science venue"
+  ],
+  "submission_ready": false,
+  "blockers": [
+    "freeze codebook and sampling rule",
+    "preregister study",
+    "execute reviewer study",
+    "human review",
+    "external replication"
+  ]
+}

research/publications/papers/2026-09-20_topology-dynamics_v0.1/CITATION.cff

@@ -0,0 +1,20 @@
+cff-version: 1.2.0
+message: "Cite this topology working paper separately from the MHRN software."
+title: "Dimensional Embedding and Propagation Dynamics in Resource-Matched Recurrent Spiking Neural Networks"
+version: "0.1"
+authors:
+  - family-names: Heisig
+    given-names: Thomas
+    orcid: "https://orcid.org/0009-0002-9589-1872"
+repository: "https://github.com/Thomas-Heisig/MHRN"
+license: MIT
+preferred-citation:
+  type: unpublished
+  status: manuscript
+  title: "Dimensional Embedding and Propagation Dynamics in Resource-Matched Recurrent Spiking Neural Networks"
+  authors:
+    - family-names: Heisig
+      given-names: Thomas
+      orcid: "https://orcid.org/0009-0002-9589-1872"
+  year: 2026
+  version: "0.1"

research/publications/papers/2026-09-20_topology-dynamics_v0.1/PAPER.md

@@ -0,0 +1,193 @@
+# Dimensional Embedding and Propagation Dynamics in Resource-Matched Recurrent Spiking Neural Networks
+
+**Thomas Heisig · Working paper v0.1 · 20 September 2026**  
+**Research strand:** topology / recurrent SNN dynamics  
+**Primary question:** `RQ-SNN-003` / `H-SNN-003-B`  
+**Status:** DATA-based working draft; human review and external replication incomplete
+
+## Abstract
+
+The geometry used to construct recurrent connectivity can alter signal propagation even when neuron count, edge budget, synaptic weight, delays, and stimulation are held fixed. We report two preregistered Stage-1 studies in the Multi-Scale Homeostatic Recurrence Network (MHRN) comparing six resource-matched topology conditions: 1D, 2D, 3D, 5D, shuffled-5D, and random-graph constructions. In the first study, 64 neurons and 246 edges were evaluated across 20 paired seeds after a separate calibration phase. First-output latency differed across all five preregistered adjacent/control contrasts after Holm correction, while terminal active fraction saturated at 1.0 for 1D–3D and therefore could not resolve those lower-dimensional differences. A prospectively corrected internal replication used new seeds and time-resolved endpoints. Activation AUC over ticks 0–32 separated 1D→2D, 2D→3D, 3D→5D, 5D→shuffled-5D, and 5D→random-graph contrasts, while all five first-output-latency directions from the original study replicated. These results support the narrow claim that concrete resource-matched topology influences propagation dynamics in this 64-neuron regime. They do not establish that a 5D embedding is generally superior, biologically privileged, more scalable, or cognitively advantageous.
+
+## 1. Introduction
+
+Recurrent spiking neural networks depend not only on neuron and synapse dynamics but also on how connectivity is arranged. A multidimensional coordinate system can be used to construct local or structured recurrent graphs, but dimensional labels alone do not imply functional advantage. A scientifically useful test must therefore separate geometric construction from confounds such as neuron count, edge density, weights, delays, stimulation, and seed choice.
+
+MHRN originally contained broader hypotheses about multidimensional address spaces. Early exploratory tests were inadequate for strong dimensional claims because they did not causally couple geometry to a sufficiently expressive recurrent graph. The present work narrows the question to a simpler one:
+
+> Under matched resource constraints, does the concrete topology used to construct a recurrent SNN alter propagation dynamics?
+
+This paper focuses on `RQ-SNN-003 / H-SNN-003-B`, not on the stronger `H-5D-005-A` advantage hypothesis.
+
+## 2. Methods
+
+### 2.1 Network design
+
+Both valid study lines use:
+
+- 64 neurons per condition;
+- 246 directed edges per condition;
+- matched synaptic weight;
+- matched delays and stimulation;
+- paired evaluation seeds;
+- separate calibration and evaluation stages where applicable.
+
+The evaluated conditions are:
+
+1. 1D structured topology;
+2. 2D structured topology;
+3. 3D structured topology;
+4. 5D structured topology;
+5. shuffled-5D control;
+6. random-graph control.
+
+The exact graph-construction rules and frozen protocol definitions are preserved in the preregistration and experiment artefacts.
+
+### 2.2 V2 preregistered study
+
+Experiment: `EXP-S1-TOPO-V2-20260918`  
+Protocol: `topology_propagation_v2`
+
+A four-seed calibration stage selected synaptic weight 55.0 through a preregistered activity-adequacy gate. The subsequent evaluation used 20 paired seeds and 120 total condition-runs.
+
+Primary endpoints were:
+
+- terminal active fraction;
+- first-output latency.
+
+The primary inferential procedure used exact paired two-sided sign tests with Holm correction across the preregistered contrast family. Deterministic bootstrap confidence intervals summarized paired median differences.
+
+### 2.3 R1 time-resolved internal replication
+
+Experiment: `EXP-S1-TOPO-V3-R1-20260918`  
+Protocol: `topology_propagation_v3_r1_time_resolved_replication`
+
+The replication used 20 new evaluation seeds and retained the 64-neuron, 246-edge, weight-55.0 design. It introduced time-resolved endpoints to resolve a known ceiling effect in terminal active fraction.
+
+Primary endpoints were:
+
+- activation AUC over ticks 0–32;
+- half-activation latency.
+
+A single Holm family covered all ten preregistered time-resolved tests. The original V2 first-output-latency contrasts were also repeated as a dedicated replication endpoint with their own Holm family.
+
+### 2.4 Evidence status
+
+Both lines are source-bound experiment DATA. They do not receive automatic EVID promotion. The R1 study is an internal preregistered replication/robustness test, not independent external replication.
+
+## 3. Results
+
+### 3.1 V2 condition summaries
+
+| Condition | n | Active fraction median | First-output latency median | Total spikes median |
+| --- | ---: | ---: | ---: | ---: |
+| 1D | 20 | 1.0000 | 19 | 401.0 |
+| 2D | 20 | 1.0000 | 9 | 363.0 |
+| 3D | 20 | 1.0000 | 6 | 281.0 |
+| 5D | 20 | 0.8750 | 5 | 181.0 |
+| 5D shuffled | 20 | 0.7734 | 2 | 179.5 |
+| Random graph | 20 | 0.7188 | 2 | 165.5 |
+
+First-output latency differed in all five primary contrasts. Median right-minus-left differences were:
+
+- 1D→2D: −10 ticks;
+- 2D→3D: −3 ticks;
+- 3D→5D: −1 tick;
+- 5D→shuffled-5D: −3 ticks;
+- 5D→random graph: −3 ticks.
+
+Each had Holm-adjusted `p = 1.90735e-05`.
+
+Terminal active fraction could not distinguish 1D→2D or 2D→3D because all three lower-dimensional conditions saturated at a median of 1.0. This is treated as a measurement-ceiling limitation rather than evidence of equality.
+
+### 3.2 R1 time-resolved results
+
+Median activation AUC over ticks 0–32 was:
+
+| Condition | Median activation AUC |
+| --- | ---: |
+| 1D | 22.1562 |
+| 2D | 26.5703 |
+| 3D | 28.0078 |
+| 5D | 25.3281 |
+| 5D shuffled | 21.0781 |
+| Random graph | 21.5078 |
+
+The five preregistered activation-AUC contrasts were all significant after correction:
+
+- 1D→2D: median Δ +4.4141, Holm `p = 1.90735e-05`;
+- 2D→3D: +1.4609, Holm `p = 1.90735e-05`;
+- 3D→5D: −2.7500, Holm `p = 1.90735e-05`;
+- 5D→shuffled-5D: −4.4141, Holm `p = 0.000160217`;
+- 5D→random graph: −3.4453, Holm `p = 1.90735e-05`.
+
+Half-activation latency separated 1D→2D and 2D→3D but not the three later contrasts.
+
+The dedicated V2 first-output-latency replication reproduced all five original contrast directions with Holm-adjusted `p = 9.53674e-06` for each contrast.
+
+The preregistered ceiling-resolution criterion and latency-replication criterion both passed.
+
+## 4. Interpretation
+
+The combined DATA support a narrow topological conclusion:
+
+> In the tested 64-neuron, 246-edge recurrent SNN regime, changing the matched graph-construction topology changes propagation dynamics.
+
+The results also show why endpoint choice matters. Terminal active fraction saturated for 1D, 2D, and 3D in V2, whereas time-resolved activation AUC in R1 resolved systematic differences among those same conditions.
+
+The observed ordering is not a monotonic dimensional-improvement result. In particular, 5D does not dominate 3D across the reported endpoints, and shuffled/random controls can produce earlier first outputs while lower activation AUC. Different propagation metrics therefore capture different graph-dynamical properties.
+
+## 5. What this paper does not show
+
+This study does **not** establish:
+
+- a general 5D advantage;
+- optimization of recurrent connectivity by dimensionality alone;
+- biological correspondence between five coordinate axes and brain anatomy;
+- scalability beyond the tested small-network regime;
+- improved learning, memory, cognition, or task performance;
+- independent external replication.
+
+The stronger 5D-specific hypotheses require a separate >=1000-neuron, geometry-specific, resource-matched programme with preregistered functional endpoints.
+
+## 6. Related work and biological context
+
+Spatial embedding, wiring constraints, recurrent graph structure, small-world organization, modularity, and conduction delay are well-established concerns in computational neuroscience and network science. Biological nervous systems exhibit strong spatial and developmental constraints, but those facts do not imply that MHRN's 5D coordinate representation maps onto biological dimensions.
+
+The appropriate biological comparison is therefore mechanistic and structural: whether controlled graph geometry produces measurable propagation effects and whether any future effect survives topology-, resource-, and scale-matched controls.
+
+A formal prior-art review is still required before external submission.
+
+## 7. Reproducibility
+
+Canonical artefacts include:
+
+- `research/preregistrations/PREREG-S1-TOPO-V2.json`;
+- `research/experiments/EXP-S1-TOPO-V2-20260918/`;
+- `research/experiments/EXP-S1-TOPO-V3-R1-20260918/`;
+- verification scripts for V2 and R1;
+- source-freeze hashes and raw evaluation DATA.
+
+The invalid intermediate V3 run is preserved as an audit trail and excluded from confirmatory interpretation because its multiple-testing implementation did not match the preregistration.
+
+## 8. Limitations and next experiments
+
+The most important next steps are:
+
+1. external replication by an independent team;
+2. larger matched networks;
+3. topology-preserving dimensional controls;
+4. resource-cost measurement;
+5. functional task endpoints beyond propagation;
+6. explicit testing of `H-5D-005-A` rather than treating `H-SNN-003-B` as a proxy.
+
+## 9. Data and code availability
+
+Code: https://github.com/Thomas-Heisig/MHRN  
+Author ORCID: https://orcid.org/0009-0002-9589-1872
+
+All numerical claims in this working draft are tied to the V2 and V3-R1 source-bound DATA artefacts identified above.
+
+## 10. Claim boundary
+
+The current manuscript is a DATA-based working paper. It may report the results of the two valid preregistered internal study lines, but it must not use phrases such as “5D is superior”, “5D optimizes connectivity”, or “brain-like dimensionality” without new, appropriately controlled evidence.

research/publications/papers/2026-09-20_topology-dynamics_v0.1/paper.json

@@ -0,0 +1,36 @@
+{
+  "schema_version": "1.0",
+  "id": "PAPER-TOPO-001",
+  "strand": "topology",
+  "title": "Dimensional Embedding and Propagation Dynamics in Resource-Matched Recurrent Spiking Neural Networks",
+  "author": "Thomas Heisig",
+  "orcid": "0009-0002-9589-1872",
+  "version": "0.1",
+  "date": "2026-09-20",
+  "status": "working_paper_data_based_review_open",
+  "research_questions": [
+    "RQ-SNN-003"
+  ],
+  "hypotheses": [
+    "H-SNN-003-B"
+  ],
+  "experiments": [
+    "EXP-S1-TOPO-V2-20260918",
+    "EXP-S1-TOPO-V3-R1-20260918"
+  ],
+  "evidence_role": "DATA_only_no_automatic_EVID",
+  "internal_replication": true,
+  "independent_replication": false,
+  "five_dimensional_advantage_claim": false,
+  "candidate_venues": [
+    "Neuromorphic Computing and Engineering",
+    "computational neuroscience / network science venue"
+  ],
+  "submission_ready": false,
+  "blockers": [
+    "human scientific review completion",
+    "formal prior-art review",
+    "external replication preferred",
+    "target journal adaptation"
+  ]
+}