Merge audited foundation reassessment for 0.0.3.7

Author: TNFR AI Agent

.github/WORKFLOWS.md

@@ -10,7 +10,6 @@ not establish current GitHub run health, branch protection or enabled secrets.
 | Workflow | Trigger and configured scope |
 | --- | --- |
 | [ci.yml](workflows/ci.yml) | Push and PR to main/master. Blocking formatting hooks, flake8 and default pytest selection on Python 3.10–3.13 with two pytest-xdist workers; combined coverage report on 3.11. Pydocstyle, mypy, pyright and vulture report advisories. |
-| [tests.yml](workflows/tests.yml) | Push and PR to main. Focused SDK tests on Python 3.11; overlaps the broader CI selection. |
 | [docs.yml](workflows/docs.yml) | Filtered push/PR to main and manual invocation. References, documentation integrity, staging and strict MkDocs build; publishes Pages only on pushes to main. |
 | [verify-references.yml](workflows/verify-references.yml) | Filtered Markdown/notebook changes on PRs and pushes to main/develop. Independent local Markdown target and fragment check. |
 | [pip-audit.yml](workflows/pip-audit.yml) | Push/PR to main/master, Monday 05:00 UTC and manual invocation. Audits the installed core, test-all and serialization environment; captures available reports and fails when the JSON audit fails. |
@@ -19,8 +18,10 @@ not establish current GitHub run health, branch protection or enabled secrets.
 | [code-review.yml](workflows/code-review.yml) | PR opened, synchronized or reopened on main. Advisory Black/mypy checks and an automated PR comment; no external code-review service is implemented. |
 | [copilot-setup-steps.yml](workflows/copilot-setup-steps.yml) | Manual invocation or edits to its own workflow. Installs a Python 3.11 development/test environment and verifies importability. |
 
-The default pytest selection excludes slow tests as configured in
-[pyproject.toml](../pyproject.toml). CI does not declare a universal coverage
+The default pytest selection is the routine engine/API gate and excludes slow
+tests as configured in [pyproject.toml](../pyproject.toml). The full retained
+research inventory is explicitly selected with `pytest tests`; it is not
+replayed by every CI job. CI does not declare a universal coverage
 percentage, prove physical theorems or certify every optional backend. Audit
 coverage excludes optional compute, documentation and deployment environments
 unless their dependencies happen to be installed through the declared groups.
@@ -64,7 +65,7 @@ invocation builds/checks only; its existing force-publish input enables publishi
 
 Publishing a GitHub release therefore also requests PyPI publication. Before
 that trigger, review validation runs for the exact commit being released:
-blocking CI formatting/static checks and its test matrix, focused SDK tests,
+blocking CI formatting/static checks and its test matrix (including SDK tests),
 the dependency audit, and the documentation/reference checks selected by that
 commit's paths. Check the workflow-invocation lint result when it is triggered.
 A skipped, cancelled or absent run is not a passing result, and success on an
@@ -77,6 +78,10 @@ verify a DOI. No GitHub/PyPI/Zenodo settings are inferred from this file.
 
 ## Retired automation
 
+The test consolidation removed `tests.yml`, which repeated the complete SDK
+suite on Python 3.11 already included in `ci.yml`. The main matrix retains the
+SDK assertions, environment checks and JUnit/coverage reporting.
+
 The second documentation cleanup removed stale workflows after preserving their
 original files under
 `artifacts/research/documentation_second_cleanup_originals_2026_09_19/`.

.github/agents/my-agent.md

@@ -1,6 +1,6 @@
 # TNFR: Resonant Fractal Nature Theory
 
-**Working reference for the TNFR Python Engine, version 0.0.3.6.**
+**Working reference for the TNFR Python Engine, version 0.0.3.7.**
 This synthesis states current definitions, contracts and mathematical boundaries.
 Guide responsibilities belong to the [documentation map](https://github.com/fermga/TNFR-Python-Engine/blob/main/docs/README.md).
 Detailed derivations belong to the [theory index](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/README.md);
@@ -29,6 +29,11 @@ that the hypothesis is physically established.
 A contradiction is resolved by examining the mathematics and actual execution;
 neither a historical label of "canonical" nor a passing test makes an unsupported
 theorem true. Update this file as a coherent reference, not a session log.
+Foundational definitions and constitutive premises are revisable hypotheses.
+State an old/new model boundary when revising them, retain valid conditional
+results, and prefer a discriminating prediction over extending a branch whose
+premises already impose its obstruction. A useful model family need not be the
+unique possible completion of the nodal identity.
 
 ### Communication policy
 
@@ -79,10 +84,24 @@ Named transformations use their operator contracts. Declared numerical solvers
 advance EPI through the shared nodal integrator with explicit capacity, pressure,
 clock and provenance or residual. Initialization is distinct from evolution.
 Full closure also needs justified phase, capacity, support and input laws.
+Validated nodal arithmetic and diagnostic scalar admission share the
+represented-real boundary: invalid types or nonzero inputs lost during
+materialization cannot certify equilibrium or a valid trajectory. Capacity
+admits zero; finite inputs still require representable output arithmetic.
+Pressure preparation and phase/capacity execution enforce that boundary before
+coercion. Temporal observers reject nonzero unrepresentable rates and retain
+detached evidence; an energy tolerance alert is not endpoint nonincrease.
+Solver clocks and retained rates use the same raw scalar admission; Boolean
+durations are not clocks. Consumed phases remain validated before caching and
+array conversion. A coordinate not consumed by a declared model need not be
+invented or coerced to execute that model.
 The optional Gamma registry supplies an additional rate in the extended row
 `dx/dt=nu_f*p+Gamma`; nonzero Gamma can move EPI at zero capacity. It is declared
 forcing, not a derivation from the nodal product. Unforced certificates require
 Gamma to be absent/zero; see [integrators](https://github.com/fermga/TNFR-Python-Engine/blob/main/src/tnfr/dynamics/integrators.py).
+The additive integrator evaluates sources strictly through the live registry;
+invalid declarations or an array fast path cannot silently remove a source.
+The opt-in extended coupled model has a separate unforced scope.
 
 ### The fractal-resonant node (NFR)
 
@@ -101,12 +120,16 @@ test does not exclude every NFR; a snapshot match or low global phase order
 does not decide the presence or absence of coherent geometric identity. See
 [research scope](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/NODAL_RESEARCH_STRATEGY.md#closure-audit-equations-events-and-identity).
 
-The radial/annular/multinodal classifier is a calibrated read-out of potential
-geometry. It does not derive a universal geometry or select Platonic solids.
-`Network.nfr()` exposes this read-out. `structural_coherence` and
+The radial/annular/multinodal classifier reads the unit-source potential
+centrality profile under a calibrated policy. Unsupported geometry is explicitly
+unavailable; a uniform profile does not prove rotational graph symmetry.
+It does not derive a universal geometry or select Platonic solids.
+`Network.nfr()` exposes this scope and the shared coherence-length estimator's
+fit/fallback provenance; it does not certify formation or fractality.
+`structural_coherence` and
 `is_structural_equilibrium` in `metrics/common.py` are shared diagnostic kernels;
-reusing them across arithmetic, graph and chemical models does not identify
-those models' dynamics. See [foundations](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/NODAL_PARAMETER_FOUNDATIONS.md).
+reusing them across graph and arithmetic models does not identify those
+models' dynamics. See [foundations](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/NODAL_PARAMETER_FOUNDATIONS.md).
 
 ### Accumulated evolution and the convergence policy
 
@@ -218,6 +241,12 @@ field energy is a Lyapunov candidate until monotonicity is proved for a stated
 law or observed on a specified trace. Read the precise hypotheses in
 [conservation](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/STRUCTURAL_CONSERVATION_THEOREM.md).
 
+Temporal balance requires nonempty matching node support and a finite positive
+interval. Each sector uses its captured graph divergence; current magnitudes
+cannot reconstruct that decomposition. Missing intervals are unavailable, not
+successful conservation. A numerical-tolerance energy alert is distinct from
+the sign of the observed energy change.
+
 Single-snapshot tensor telemetry reports conservation quality as unavailable;
 node-label ordering cannot supply a time derivative or graph divergence.
 Composite read-outs reuse one base-field collection and retain its node order.
@@ -334,6 +363,18 @@ that dependency is not an emergent selection law. The auxiliary affinity
 availability, stored versus refreshed pressure and xi fit/fallback provenance.
 Do not treat undefined or unsupported observations as successful zero values.
 
+Automatic candidates use admitted stored metrics, current normalization and an
+explicit batch lifetime; operator admission remains independent. Configured
+phase labels depend on sample size and cannot establish autonomous formation.
+An incomplete tetrad cannot certify even the configured overall safety advisory.
+
+Thresholds must retain their units and actual consumer: normalized selector
+pressure is distinct from absolute adaptation pressure. Shared default
+resolution must agree across selection, adaptation and diagnostics. A pi-based
+formula or a legacy `CANONICAL` name does not derive a policy; numerical
+tolerances and inactive compatibility exports supply no physical theorem.
+See the [threshold ledger](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/NODAL_PARAMETER_FOUNDATIONS.md#11-thresholds-constants-and-numerical-settings-have-different-duties).
+
 ## 8. Canonical invariants
 
 1. Preserve continuous nodal flow and declared hybrid jumps with their provenance.
@@ -420,6 +461,9 @@ for the tests that retain the behavior of retired introductory illustrations.
 Default test selection, dependencies and parallel scheduling have their owners
 in test configuration and the workflow guide; do not duplicate those settings
 here or regenerate historical research artifacts for an unrelated change.
+The default run is the routine engine/API gate, not the full research inventory.
+Select the affected research owner explicitly when changing its model or claim;
+report that scope rather than presenting a routine pass as complete coverage.
 See [TESTING.md](https://github.com/fermga/TNFR-Python-Engine/blob/main/TESTING.md).
 
 ## 11. Troubleshooting
@@ -438,8 +482,10 @@ classifies primary, supporting and parked work. The
 is the sole task queue; the
 [strategy](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/NODAL_RESEARCH_STRATEGY.md)
 explains its rationale. Conditional core results, finite C6 evidence, arithmetic
-models and Millennium notes retain their own scope and do not constitute
-additional active campaigns. Historical instructions are preserved in the
+models and scoped auxiliary algebra retain their own hypotheses and do not
+constitute additional active campaigns. Unsupported physical-programme
+wrappers and circular type probes are retired; their boundaries and intentional
+API removals are recorded in the
 [archive](https://github.com/fermga/TNFR-Python-Engine/blob/main/theory/research/archive/README.md).
 
 The primary objective is a predictive generative account of coherence patterns.

.github/workflows/ci.yml

@@ -89,7 +89,7 @@ jobs:
         continue-on-error: true
 
   tests:
-    name: Python ${{ matrix.python-version }} tests
+    name: Python ${{ matrix.python-version }} engine tests
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
@@ -117,11 +117,11 @@ jobs:
         continue-on-error: true
         run: python -m pydocstyle src/tnfr
 
-      - name: Run unit tests
+      - name: Run routine engine and API tests
         if: matrix.python-version != '3.11'
         run: python -m pytest -n 2 --dist loadfile --junitxml=test-results/pytest.xml
 
-      - name: Run unit tests with coverage
+      - name: Run routine engine and API tests with coverage
         if: matrix.python-version == '3.11'
         run: >-
           python -m pytest -n 2 --dist loadfile

.github/workflows/tests.yml

@@ -1,8 +1,8 @@
 {
   "title": "TNFR-Python-Engine: Resonant Fractal Nature Theory Implementation",
-  "description": "<p>TNFR-Python-Engine is a Python framework for coherent-pattern analysis on graph-coupled networks. The nodal relation dEPI/dt = nu_f DeltaNFR organizes a declared evolution model; pressure, phase, capacity, support and any additional forcing require their own specifications.</p><p>The engine provides 13 registered structural operators, grammar U1-U6, shared integration and the structural-field tetrad: potential, phase-gradient magnitude, circular phase curvature and coherence length. These fields are diagnostics; they do not reconstruct every nodal state or independently select its evolution. Operator contracts distinguish primary effects, auxiliary writes, grammar admission and execution provenance.</p><p>Version 0.0.3.6 integrates scoped diffusion, memory, phase/form and quotient research with corrected documentation. Exact identities, conditional theorems, represented finite observations, auxiliary wave/Hamiltonian/arithmetic models and open hypotheses are identified separately. Passing tests is not proof of unrestricted stability, autonomous physical emergence or a Millennium conjecture. The research execution plan remains the single source of current priorities.</p><p>The shared SDK study API and CLI construct and execute declared operator-word studies through the existing engine. JSON recipes and reports retain supplied initial state, software versions, diagnostic availability and coherence-length provenance. Recipe cycles are operator-word counts, not physical seconds, and recipes are not complete checkpoints or cross-version replay guarantees.</p><p>The release consolidates technical guides and test helpers, removes the retired educational manuscript and obsolete standalone demonstrations, reuses existing numerical evaluations, and validates documentation against implemented contracts. Scoped regression checks and historical research evidence remain distinct from retired toy tests and heuristic campaigns. See the <a href=\"https://github.com/fermga/TNFR-Python-Engine/releases/tag/v0.0.3.6\">0.0.3.6 release</a>, <a href=\"https://github.com/fermga/TNFR-Python-Engine/tree/v0.0.3.6\">versioned source</a>, <a href=\"https://pypi.org/project/tnfr/0.0.3.6/\">Python package</a> and <a href=\"https://fermga.github.io/TNFR-Python-Engine/\">current documentation</a>.</p>",
-  "version": "0.0.3.6",
-  "publication_date": "2026-09-19",
+  "description": "<p>TNFR-Python-Engine is a Python framework for coherent-pattern analysis on graph-coupled networks. The nodal relation dEPI/dt = nu_f DeltaNFR organizes a declared evolution model; pressure, phase, capacity, support and any additional forcing require their own specifications.</p><p>Version 0.0.3.7 consolidates shared numerical admission, operator contracts, cache invalidation and runtime execution provenance. Runtime step ordinals are independent of retained metrics and physical time. REMESH uses explicit cooldown bases and bounded histories. SDK and CLI reports retain unavailable diagnostic results and provenance instead of substituting evidence. The 13 registered operators, grammar U1-U6, shared integration and structural-field tetrad remain scoped engine interfaces.</p><p>The release revises foundational documentation and centralizes the active research plan. A fixed finite P3 experiment compares two declared phase-pressure laws against independent form-response predictions, with pressure, integration and rounding defects separated. It distinguishes their implemented responses without selecting a physical law or proving autonomous pattern formation. Exact identities, conditional theorems, numerical observations and auxiliary models retain separate scopes.</p><p>Unsupported particle and chemistry mappings, Navier-Stokes and Yang-Mills wrappers, redundant type probes and obsolete demonstrations are removed. These are intentional API removals; migration boundaries and recoverable historical sources are documented in the archive index. Routine engine/API checks are separated from retained research suites, which remain explicitly executable.</p><p>See the <a href=\"https://github.com/fermga/TNFR-Python-Engine/releases/tag/v0.0.3.7\">0.0.3.7 release</a>, <a href=\"https://github.com/fermga/TNFR-Python-Engine/tree/v0.0.3.7\">versioned source</a>, <a href=\"https://pypi.org/project/tnfr/0.0.3.7/\">Python package</a>, <a href=\"https://github.com/fermga/TNFR-Python-Engine/blob/v0.0.3.7/theory/research/archive/README.md\">compatibility boundaries</a> and <a href=\"https://fermga.github.io/TNFR-Python-Engine/\">current documentation</a>.</p>",
+  "version": "0.0.3.7",
+  "publication_date": "2026-09-20",
   "creators": [
     {
       "name": "Martinez Gamo, F. F.",
@@ -33,7 +33,7 @@
       "resource_type": "software"
     },
     {
-      "identifier": "https://pypi.org/project/tnfr/0.0.3.6/",
+      "identifier": "https://pypi.org/project/tnfr/0.0.3.7/",
       "relation": "isSupplementTo",
       "resource_type": "software"
     }