Sync GitHub main 466ae6e87de276549cff5a7ef52c55419843ace7

.github/workflows/ci.yml

@@ -7,7 +7,7 @@
 #   1. lint-format   — code style, linting, pre-commit
 #   2. type-check    — static type analysis (mypy + pyright)
 #   3. security      — SAST / dependency scanning
-#   4. tests         — full test suite across Python versions (matrix)
+#   4. tests         — full test suite on the canonical Python 3.13 CI runtime
 #   5. build         — wheel build + install verification
 #   6. docker        — Docker image build test
 #   7. docs          — documentation structure check
@@ -243,7 +243,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ["3.11", "3.12", "3.13"]
+        python-version: ["3.13"]
     steps:
       - uses: actions/checkout@v7
         with:

.zenodo.json

@@ -32,5 +32,5 @@
       "resource_type": "other"
     }
   ],
-  "version": "0.6.0-alpha.7"
+  "version": "0.6.0-alpha.8"
 }

CITATION.cff

@@ -3,7 +3,7 @@ message: "Cite the MHRN software and the Recursive Epistemics publication separa
 title: "Multi-Scale Homeostatic Recurrence Network (MHRN)"
 abstract: "Mehrskaliges homöostatisches Rekurrenznetzwerk. A Spiking Neural Architecture with Topological Plasticity."
 type: software
-version: "0.6.0a7"
+version: "0.6.0a8"
 authors:
   - family-names: Heisig
     given-names: Thomas

HF_README.md

@@ -156,7 +156,7 @@ MIT License — see `LICENSE`.
   author  = {Thomas Heisig},
   title   = {Multi-Scale Homeostatic Recurrence Network (MHRN)},
   year    = {2026},
-  version = {0.6.0a7},
+  version = {0.6.0a8},
   url     = {https://github.com/Thomas-Heisig/MHRN},
   license = {MIT}
 }

MIRROR_MANIFEST.json

@@ -1,6 +1,6 @@
 {
   "canonical_repository": "https://github.com/Thomas-Heisig/MHRN",
-  "canonical_sha": "4a1051375bd481051c1e3c999d5e5e0679f2bf7e",
+  "canonical_sha": "466ae6e87de276549cff5a7ef52c55419843ace7",
   "generated_by": ".github/workflows/sync-huggingface.yml",
   "rolling_mirror": true,
   "raw_experiment_data_included": false,

OPEN_SCIENCE.md

@@ -20,7 +20,7 @@ This file is the public routing map for MHRN research visibility. GitHub is the
 | Hugging Face MHRN-Research-Data | rolling research-tree discovery mirror | **CONFIGURED** | same credentials; workflow creates and synchronizes `MHRN-Research-Data`; immutable experiment DOIs remain separate |
 | ORCID | author identity and cross-service identity anchor | **ACTIVE in repository** — `0009-0002-9589-1872` | verify public ORCID record and authorize trusted auto-updates where desired |
 | OSF | project landing page; registrations/preregistrations | **ACTIVE in repository** — `https://osf.io/p34uq/` | use registrations for protocol freezes; do not rely on OSF Projects as the only long-term file store |
-| Zenodo | immutable software/data/preprint deposits and DOI minting | **CONFIGURED** — `.zenodo.json` + `CITATION.cff` present | enable/verify MHRN in Zenodo GitHub integration and publish DOI-bearing records |
+| Zenodo | immutable software/data/preprint deposits and DOI minting | **ACTIVE** — Alpha.7 software archive published and recorded at [10.5281/zenodo.23138847](https://doi.org/10.5281/zenodo.23138847) | create separate DOI deposits for release-quality experiment datasets and suitable preprints; keep each record linked to its exact source/data freeze |
 | DataCite | DOI metadata network | **HARVEST TARGET** | reached through Zenodo and other DOI repositories; ensure ORCID and relations are present in DOI metadata |
 | OpenAIRE Research Graph | research-object aggregation and linking | **HARVEST TARGET** | verify records after DOI publication; link funding/project metadata only when factual |
 | OpenAlex | scholarly graph / DOI and author discovery | **HARVEST TARGET** | verify DOI/ORCID records after indexing |
@@ -116,8 +116,8 @@ For each release-quality scientific object:
 
 Repository metadata prepares the records, but these require the account owner or an external submission system:
 
-- enable/verify the Zenodo-GitHub repository connection;
-- publish Zenodo deposits/DOIs;
+- verify the Zenodo-ORCID account link and authorize DOI auto-updates where desired;
+- publish separate Zenodo deposits/DOIs for release-quality experiment datasets and suitable preprints;
 - create/finalize OSF registrations;
 - authorize ORCID update permissions;
 - verify/claim OpenAIRE, OpenAlex, Semantic Scholar, Google Scholar, HAL or ResearchGate records where user action is offered;

README.md

@@ -156,7 +156,7 @@ MIT License — see `LICENSE`.
   author  = {Thomas Heisig},
   title   = {Multi-Scale Homeostatic Recurrence Network (MHRN)},
   year    = {2026},
-  version = {0.6.0a7},
+  version = {0.6.0a8},
   url     = {https://github.com/Thomas-Heisig/MHRN},
   license = {MIT}
 }

codemeta.json

@@ -6,7 +6,7 @@
   "description": "Experimental research framework for recurrent spiking neural networks, plasticity, self-organization, embodiment, memory and world-model mechanisms under explicit provenance, reproducibility and evidence boundaries.",
   "codeRepository": "https://github.com/Thomas-Heisig/MHRN",
   "url": "https://github.com/Thomas-Heisig/MHRN",
-  "version": "0.6.0a7",
+  "version": "0.6.0a8",
   "license": "https://spdx.org/licenses/MIT",
   "programmingLanguage": [
     "Python",

docs/07-changelog/CHANGELOG.md

@@ -1,5 +1,24 @@
 # Changelog
 
+## 2026-10-17 - Alpha.8 release candidate preparation (pre-publication)
+
+- Restored green `develop` CI (strict pyright typing in the raw-run index) and
+  merged integration PR #282 at `343d1634`, including the `OPEN_SCIENCE.md`
+  Zenodo alignment and the manuscript-review follow-up TODOs.
+- Bumped the package to `0.6.0a8` and added alpha.8 candidate release notes
+  and release records. Alpha.8 is an integration/maintenance pre-release; the
+  legacy "Alpha.8" recursive-loopback programme is not included.
+- No DATA, Human Review, EVID, scientific maturity or replication state
+  changed. Tag, GitHub release and Zenodo DOI are pending their gates.
+
+## 2026-10-04 - Bounded cognition raw-run viewer
+
+- Added File Viewer drill-down from a registered cognition `runs_index.json` to
+  immutable compressed raw runs. The viewer exposes at most 100 indexed runs;
+  previews require a matching SHA-256 entry and decompress at most 256 KiB.
+- Raw runs remain read-only DATA. This viewer does not create analysis, Human
+  Review, EVID or scientific status promotion.
+
 ## 2026-10-04 - Alpha.7 pre-release and Zenodo archive verified
 
 - Published GitHub pre-release `v0.6.0-alpha.7` on source freeze
@@ -14,15 +33,28 @@
 - Publication and DOI assignment do not change DATA, Human Review, EVID,
   scientific maturity or independent-replication state.
 
-## 2026-10-04 - Alpha.7 release candidate integration status
+## 2026-10-04 - Develop synchronization and CI runtime
+
+- Integrated released `main` into `develop` without rewriting or dropping
+  develop-only work.
+- Reduced the full-suite CI matrix to Python 3.13. Focused security, typing,
+  contracts, browser, documentation, build and Docker gates remain enabled.
+  Package metadata still declares `>=3.11`; CI no longer exercises 3.11/3.12.
+- Published the existing relative-path Pages fix with review-portal run
+  `37214631300` on `main@4a105137`; root, stylesheet, app and admin routes
+  returned HTTP 200.
+- Restored the develop lint gate by formatting the Stage-4 integration helper
+  and fixing a stale benchmark-test helper name; no runtime behavior changed.
+
+## 2026-10-04 - Alpha.7 release candidate preparation (pre-publication)
 
 - Merged integration PR #269 to `main` at `4a6eb988`, including the publication
   synthesis and Stage-3 benchmark diagnostics. Scientific status, Human Review,
   EVID and replication state were not promoted.
 - Added Alpha.7 candidate release notes and reconciled the release ledgers to
-  the merged integration. The dedicated release PR and post-merge CI/source
-  freeze remain pending; no tag, GitHub release or Alpha.7 Zenodo archive has
-  been created.
+  the merged integration. At that preparation point the dedicated release PR,
+  post-merge CI, tag and Zenodo archive were pending; the publication entry
+  above records their later completion.
 - The configured Hugging Face sync succeeded at the integration commit.
   Zenodo requires the eventual GitHub release; OSF and ORCID have no configured
   automatic release-update workflow, and no institution-specific endpoint is
@@ -37,7 +69,16 @@
 - Re-ran 10k/100k for all three modes and heterogeneous asymmetric 100k/1M.
   The uniform-input all-zero case is retained as a separate symmetric workload,
   not treated as a general implementation failure. New schema-v2 receipts are
-  separate from earlier receipts.
+  separate from earlier receipts and include benchmark-source/topology SHA-256.
+- The uniform 10k/100k rerun confirms all weights hit zero; numeric bounds pass
+  but the functional stability gate fails. A seed-aware 42/43/44 by
+  asymmetric/symmetric/off sweep was also recorded; its engineering variation
+  is not independent biological replication.
+- In matched heterogeneous 100-epoch runs, asymmetric STDP increases lower-bound
+  occupancy by 12.5 percentage points relative to symmetric STDP, while
+  symmetric STDP also saturates the upper bound. This is consistent with a
+  depression contribution but does not isolate a sole cause or a universal
+  parameter recommendation.
 - Added verified STDP, homeostasis, memory, continual-learning, sensorimotor,
   network-organization, preregistration, and simulator-comparability sources,
   mapped to the relevant research areas. No direct 5D-advantage precedent was
@@ -75,10 +116,12 @@
 - Confirmed that the seven registered Gateway protocols are boundary audits,
   not executable Plastic protocols; their audit seed reexecutions do not count
   as independent Plastic runs.
-- The current Gateway route accepts caller-supplied preregistration JSON and
-  activates a single seed. No registered Plastic runner, frozen input-frame
-  sequence or executable stop rule is available, so registry-backed
-  multi-seed execution remains open rather than inventing a workload.
+- The HTTP Plastic activation route now resolves protocol IDs against the
+  operational registry and rejects unknown or boundary-audit protocols. It
+  remains fail-closed even for a future functional protocol until a registered
+  multi-seed runner exists; it does not execute one caller-supplied seed.
+- No frozen functional Plastic runner, input-frame sequence or executable stop
+  rule is available, so registry-backed multi-seed execution remains open.
 - No scientific DATA/EVID status is inferred from Gateway activation or audit
   artifacts.
 

docs/08-roadmap/ROADMAP.md

@@ -1,21 +1,37 @@
 # MHRN Development Roadmap
 
 **Canonical roadmap for active `develop`; `main` is release-only**  
-**Baseline:** `mhrn-core 0.6.0a7`
-**Updated:** 2026-10-04
+**Baseline:** `mhrn-core 0.6.0a8`
+**Updated:** 2026-10-17
+
+## 2026-10-17 Alpha.8 release line
+
+- `v0.6.0-alpha.8` is cut from green `develop@343d1634` (PR #282) through
+	`release/v0.6.0-alpha.8`. It is an integration/maintenance pre-release
+	without scientific status changes.
+- The recursive causal loopback programme (legacy alpha.8 TODO label) remains
+	open and is not part of this release.
+
+## 2026-10-04 Cognition raw-run File Viewer
+
+- Added run-level drill-down from a registered cognition `runs_index.json`.
+	Only indexed research artifacts with a matching SHA-256 are previewed; the UI
+	lists at most 100 runs and backend decompression is capped at 256 KiB.
+- Source artifacts remain immutable DATA. Preview access does not constitute
+	analysis, Human Review, EVID or scientific status promotion.
 
 ## 2026-10-04 Gateway Plastic preregistration and execution boundary
 
 - The seven registered `RQ-GW-*` operational protocols are `boundary_audit`
   contracts with `direct_test_of_hypothesis=false`; their seed lists describe
   audit reexecution, not independent Gateway Plastic runs.
-- The Gateway activation route accepts caller-provided preregistration JSON and
-  starts one runtime seed. There is no registered functional Plastic runner,
-  frozen input-frame sequence, or executable stop rule to define a reproducible
-  trial across independent runtime instances.
-- Keep registry-backed Plastic activation and multi-seed execution open until
-  a suitable frozen functional protocol and runner contract exist. No Gateway
-  DATA/EVID is inferred from the current audit protocols or activation status.
+- The public Plastic activation route resolves `protocol_id` against the
+	operational registry and rejects unknown or `boundary_audit` entries before
+	activation. A single-seed API call cannot authorize a Plastic run.
+- The seven registered Gateway protocols remain boundary audits. No frozen
+	functional Plastic protocol, input-frame sequence, executable stop rule or
+	independent multi-seed runner exists yet; keep Plastic execution and Gateway
+	DATA/EVID blocked until those contracts are independently reviewable.
 
 ## 2026-10-04 Stage-3 plastic-network scale benchmark
 
@@ -26,7 +42,9 @@
 	boundedness is reported separately from ongoing spikes, active-weight
 	fraction, and workload-required weight diversity. Candidate visits/s is an
 	operation-count estimate, not a hardware counter. Receipts include workload,
-	software and CPU metadata; all remain engineering-only.
+	software, CPU, benchmark-source SHA-256 and topology SHA-256; all remain
+	engineering-only. Pair-STDP, eligibility, and reward modulation are included;
+	homeostasis, structural growth/pruning, and checkpoint/restore are excluded.
 - Matched 10,000-neuron / 100,000-synapse runs (seed 42, 100 epochs): the
 	heterogeneous asymmetric profile ended with 56.25% of weights at the lower
 	bound, 43.75% active and positive variance; the symmetric profile ended with
@@ -34,9 +52,21 @@
 	profile retained all weights at 0.05. The earlier uniform-input profile drove
 	all weights to zero. Its identical weights are expected under that symmetric
 	input/topology; it is not evidence by itself of an update implementation bug.
+- In matched heterogeneous 100-epoch runs, asymmetric `a_minus > a_plus` has
+	12.5 percentage points more lower-bound weights than symmetric STDP, which is
+	consistent with a depression contribution. Symmetric STDP also puts 18.75%
+	at the upper bound, so the comparison does not isolate a sole collapse cause
+	or establish a generally preferred parameterization.
+- Repeating that uniform-input profile with V2 checks reproduces 100% lower-bound
+	saturation and correctly reports `numeric_stability_passed=true`, but
+	`functional_activity_passed=false` and `stability_invariants_passed=false`.
+- The `seed` now deterministically permutes target-node order. A 42/43/44,
+	three-mode, 20-epoch sweep at 10k/100k produced active fractions of 50%
+	(asymmetric), 62.5% (symmetric), and 100% (off) for every seed. This checks
+	engineering topology sensitivity only; the seeds are not biological samples.
 - At 100,000 neurons / 1M synapses for 20 heterogeneous asymmetric epochs, the
 	measured Python peak was about 1.03 GB, the estimated candidate-visit rate
-	about 210k/s, and 50% of weights remained active. These workload-specific
+	about 195k/s, and 50% of weights remained active. These workload-specific
 	profiles do not establish general or long-horizon stability.
 - The declared 100,000-neuron / 10M-synapse upper point remains unmeasured:
 	its preflight estimate is 19.45 GiB, exceeding this machine's available RAM.
@@ -137,6 +167,20 @@
 - Release and DOI publication leave Human Review, EVID, maturity and independent
 	replication state unchanged.
 
+## 2026-10-04 Develop continuation
+
+- Released `main@4a105137` is integrated into `develop` by merge commit
+	`97fa42db`; the four develop-only commits remain in history.
+- Full-suite CI now runs on Python 3.13 only. The other security, typing,
+	contract, browser, documentation, build and Docker gates remain enabled.
+	`pyproject.toml` still declares Python `>=3.11`; 3.11/3.12 are no longer CI
+	tested.
+- The export-only GitHub Pages portal was republished by workflow run
+	`37214631300` from `main@4a105137`. Root, stylesheet, application module and
+	admin page all returned HTTP 200 in the live smoke check.
+- A develop-only formatting drift and stale benchmark-test helper name were
+	corrected; the fix changes no runtime behavior.
+
 ## 2026-09-29 Post-hoc Experimentauswertung
 
 - Experimentkarten im Wissenschaftsbereich erlauben die nachträgliche