Consolidate pair-state research, shared contracts and repository cleanup

Integrate the full unordered pair-state representation, cancellation observations,
finite-response and retained-pattern research with shared admission and SDK reports.
Keep native relational execution and conditional sine comparison laws distinct.

Split oversized theory and API documents by responsibility, retain canonical
anchors and frozen evidence, and centralize the resumption queue. Retire unused
helpers and superseded studies with explicit migrations and recovery records.

Resolve structural-validation configuration collisions and inert settings; check
actual README code and strict publication JSON. Install declared core dependencies
for the metadata gate, preserve zero partition overlap, and keep parallel test
reports serializable without weakening their assertions.

Validation: routine engine/API selection reached 9256 passed and 24 skipped;
the sole failure was xdist serialization of a NumPy subtest label, corrected and
revalidated (8 tests and 2 subtests passed). Zero-overlap configuration and
factorization export checks also passed (8 tests). Changed research contracts, imports,
exports, formatting, static checks, documentation links/site, package parity and
isolated publication metadata were checked separately. This is conditional
mathematical/computational evidence, not physical validation or a replay of every
retained research campaign. Detailed local logs are in
artifacts/final_audit_publish_20261007/.

Author: TNFR AI Agent

.github/WORKFLOWS.md

@@ -54,7 +54,9 @@ Edit source owners, not generated site copies.
 ## Package publication
 
 The retained publisher is [pypi-zenodo.yml](workflows/pypi-zenodo.yml); its filename
-is historical. It first checks publication metadata and, for a release event,
+is historical. It installs the checkout and its declared core dependencies with
+the build tools so metadata validation can use the shared JSON reader. It then
+checks publication metadata and, for a release event,
 requires the tag to match the declared package version. It then performs build
 and Twine checks and requests PyPI trusted publishing. It neither creates a
 release nor runs the numerical test suite.
@@ -75,35 +77,14 @@ The final Zenodo step only prints a reminder. A separately configured external
 integration may archive a release; this workflow does not upload an archive or
 verify a DOI. No GitHub/PyPI/Zenodo settings are inferred from this file.
 
-## Retired automation
-
-The former `code-review.yml` job and reminder-only pre-commit hook are retired.
-Black and advisory mypy already have maintained owners in `ci.yml`; repeating
-them in another environment and posting an automatic PR comment added no
-independent review coverage. The main CI gates retain their existing scope.
-
-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/`.
-The reasons are recorded here to prevent restoring broken obligations:
-
-- The former push-to-main release workflow duplicated PyPI publication and
-  depended on absent changelog configuration, fragment checker and test script.
-  The missing changelog job was also removed from CI.
-- The performance guard targeted an absent performance-test directory.
-- The tetrad smoke workflow invoked an absent runner.
-- The manual reproducibility workflow targeted an absent profiler and manifest
-  comparison script.
-
-Retained CI provides the stated checks; it is not equivalent replacement
-coverage for those retired benchmark protocols. Research checks use their
-maintained, explicitly documented entry points in
-[benchmarks/README.md](../benchmarks/README.md). There is no automatic benchmark
-or reproducibility guarantee implied by an ordinary CI pass.
+## Research verification boundary
+
+Ordinary CI exercises the declared engine/API selection. Research campaigns
+use the explicitly selected owners in [Testing](../TESTING.md) and
+[benchmarks](../benchmarks/README.md). A routine pass is not an automatic
+benchmark, reproducibility certificate or replay of every retained experiment.
+Retirement and recovery records belong to the
+[archive](../theory/research/archive/README.md), not this workflow inventory.
 
 ## Maintenance rules
 

.github/agents/my-agent.md

@@ -28,10 +28,10 @@ jobs:
       with:
         python-version: '3.11'
 
-    - name: Install build dependencies
+    - name: Install build and metadata-validation dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install "build>=1,<2" "twine>=6,<8"
+        python -m pip install . "build>=1,<2" "twine>=6,<8"
     
     - name: Verify publication metadata and release tag
       env:

.github/workflows/pypi-zenodo.yml

@@ -4,6 +4,191 @@ All notable changes to this project will be documented in this file.
 
 ## [Unreleased]
 
+- Retain the full unordered pair state through phase cancellation in the
+  conservative unit doubled-C5 sine model. Add exact realizability admission,
+  inherited polynomial rates, current derivatives and storage observations
+  with SDK export. The representation needs no midpoint angle, new law or
+  extra continuous degrees of freedom; existing regular adapters keep their
+  own domains.
+
+- Consolidate the retained moving-pair interaction contract from existing
+  replica, current and memory results. Distinguish constant collective means
+  from internally modulated compensating currents, and present messages from
+  sufficient future state. Regional-transfer and form-increment reports now
+  attach a rebuilt comparison, preventing stale nested evidence from contradicting
+  the freshly computed ledger.
+
+- Connect phase/rate observations to the existing unordered pair state through
+  an exact regular three-node star chart. Recover relative form and inherited
+  evolution, reject singular observation domains, and retain explicit coincident
+  and antipodal counterexamples. This changes representation without selecting
+  a law or removing continuous degrees of freedom.
+
+- Consolidate phase information and its motion through shared exact phasor
+  admission, harmonic moments and source/storage calculations. Add a derived
+  phase/rate association observer with an explicit full-star counterexample;
+  no new free parameter or reduced evolution law is introduced. Correct the
+  distinction between odd currents and cancellation under a half turn.
+- Rebuild sine comparison, mobility and environmental-mediation calculations
+  from admitted primitive state before consuming gradients, rates or work.
+  Altered cached values no longer change those derived results; invalid
+  primitive state remains rejected.
+- Apply the same report-consumption boundary to joint pairing windows and
+  structural Emission increments. Reuse shared source/window and AL proposal
+  builders so cached bounds, target indices and increments cannot become
+  independent dynamical premises.
+
+- Add finite bridge-response discriminators for the sine and one-percent
+  cubic-current laws. Opposite form preparations and an independent phase
+  preparation give exact finite-amplitude curvature coefficients. Shared
+  full-state bounds retain preparation, reading and time-remainder errors;
+  complete endpoint budgets also certify a ratio under a common unknown clock
+  scale within ten percent. These analytical predictions neither install a
+  law nor supply a physical measurement.
+
+- Resolve the fixed one-percent cubic-current comparison by a conserved-storage
+  obstruction: the unchanged mapped source cannot acquire either acute unit
+  winding under that law, although its critical geometries remain available.
+  Add a read-only evidence audit without replaying or replacing the frozen
+  sine response; distinguish energetic accessibility from equilibrium stability.
+
+- Add a complete smooth phase-storage countermodel with exact local agreement
+  and a different global winding barrier. Integrate a static full-state
+  discriminator and SDK projection, retaining uncertainty, both time directions
+  and the distinction between smooth and real-analytic law classes.
+
+- Add a fresh two-direction conservative connection evaluator and freeze its
+  declaration, full response and source. Certify formation and finite acute
+  retention on validated prefixes while preserving the incomplete forward
+  horizon and the separate independent-source obligation.
+- Reuse cached trigonometric Taylor prefixes without changing recurrence
+  arithmetic; verify nonlinear composition and cache eviction against exact
+  analytic coefficients. Preserve the earlier frozen calculation's source.
+
+- Certify metric growth from the actual whole-tube Jacobian using exact
+  rational matrix inequalities and a fixed rate bracket. Reuse signed
+  derivative admission across comparison solvers; retain original-time
+  rates and explicit failures beyond a declared numerical budget.
+
+- Add a shared Picard/Taylor step retaining full-state metric uncertainty
+  between steps, with complete local rounding/remainder bounds. Integrate
+  the conservative C5 saddle adapter, original-clock forward/reverse rates,
+  strict whole-time phase admission and SDK export. Keep partial coverage
+  explicit and distinguish solver verification from a formation experiment.
+
+- Admit explicit rational near-saddle preparations with retained representation
+  error, resolved near-critical storage and both local directional gates.
+  Centralize full and regional sine storage on the shared 128-bit interval
+  cosine. Add an exact full-state sensitivity metric and conditional nonlinear
+  phase-tube growth bound, without claiming a validated metric forecast or an
+  available zero-winding source.
+
+- Prove a same-orbit conservative formation-to-retention construction on
+  C5/private-leaf support, including independent full-state source and target
+  widths and one scaled unit of acute retention. Reuse nonlinear corridors
+  and a phase-band dwell bound in shared analytical reports and SDK exports;
+  keep event-defined centers unavailable and the tiny positive source radius
+  symbolic. Separate this existence result from operational preparation.
+
+- Add a full-law conservative saddle-corridor certificate: retained receiver
+  and leaf momentum can select a finite principal-seam passage without a
+  trajectory or tangent extrapolation. Expose the conditional result through
+  shared SDK reports; acute retention and transverse uncertainty remain
+  separate obligations. Extend the cycle-sector obstruction to the exact
+  barrier energy using boundary equilibrium and flow uniqueness.
+
+### Conditional dynamics and shared evidence
+
+- Integrate conservative bridge memory and collective observations with
+  explicit hidden initialization, channel distinctions and source admission.
+  Add form-loss observation obstructions without identifying apparent local
+  attenuation with microscopic dissipation.
+- Extend phase/storage information and inverse-geometry controls, preserving
+  constitutive freedom and independently frozen response evaluation.
+- Add retained-frame composition, conservative regional entry/work observers,
+  invariant phase-offset families, finite moving-pattern retention and actual
+  collective-pulse feedback. Formation, maintenance and physical identification
+  retain separate hypotheses; conservative transfer does not select support
+  birth or autonomous preparation.
+- Preserve positive witnesses, handoff obstructions and frozen negative
+  controls through shared validated flow, exact report projection and SDK
+  export. Correlated cycle storage sharpens target-budget admission; the
+  collective pulse's signed storage split retains its phase-lift dependence.
+- Add source admission from the conservative C5 sector barrier and a finite
+  rapid-contact averaging bound. The former includes nonflat zero-winding
+  sources; the latter can exclude acute formation despite large storage.
+  Neither passing an admission nor preserving the wider sector proves acute
+  formation, acute retention or physical identity.
+- Extend the existing conservative entry certificate with full-state source
+  uncertainty and an explicit independently continuable entry box. Separate
+  nominal storage from uncertain source storage and robust short passage
+  from the stronger research requirement of extended finite retention.
+- Resolve receiver phase velocity, acceleration and third derivative through
+  the existing collective-pulse observer. Retain environmental feedback and
+  document the conditional rigid-motion classification and a compensated
+  initial state whose relative phase motion reappears at third order.
+- Add a full-state C5 retention window from conserved storage and actual
+  full-support phase-speed bounds. It covers uncertain, internally moving
+  targets without an invariant reference; prepared retention and acquisition
+  from zero winding remain separate claims.
+- Add a shared reversible-preparation certificate that retains inverse-flow
+  enclosure error and an independent source radius in a strict forward
+  inclusion bound. Reuse the existing frozen producer with a separately
+  admitted declaration mode; preserve historical response schemas and defaults.
+- Add exact signed-involution reconstruction and a conservative C5/private-leaf
+  saddle assessment. Reuse full support admission and tangent assembly;
+  distinguish invariant families, captured-source membership, local unstable
+  directions and nonlinear passage from actual winding acquisition.
+
+### Consolidated documentation
+
+- Align the introduction, foundations, architecture and public contracts with
+  the implemented mechanisms and their limitations. Connect pulse, resonance,
+  retained memory and scale inheritance without transferring results between
+  incompatible laws or supports.
+- Keep the execution plan as the sole task queue, the portfolio as branch
+  classification and the strategy as rationale. Frozen evidence and historical
+  source identifiers remain preserved; no new release is implied by this entry.
+
+### Cleanup and interface corrections
+
+- Remove five ineffective global structural-validation settings and normalize
+  the consumed flags/severity through shared parsers. Reject unknown or invalid
+  update batches before committing any setting. Expose that policy as
+  `StructuralValidationConfig` without changing the existing unified-validator
+  `ValidationConfig` export, and align runtime and stub declarations. Point the
+  resumption gate directly to its retained pair-state and observation contracts.
+- Execute the actual README quick start in documentation checks and reject
+  duplicate keys in publication metadata through shared JSON admission. Check
+  finite-work responses against independently constructed doubled-C5 support.
+  Replace two sine-recovery forwarding wrappers with direct shared-owner aliases.
+- Install declared core dependencies in the publication workflow before running
+  the shared metadata validator; retain one dependency source in package metadata.
+- Preserve explicitly zero partition overlap in the optional factorization lab
+  instead of silently creating overlapping partitions. Keep test singleton
+  configuration isolated and classical-adapter subtest labels serializable for
+  parallel execution.
+- Remove unused parallel mathematical score/field helpers, duplicate registry
+  observations and grammar-cache wrappers. Preserve shared nodal calculus,
+  graph fields, live grammar and class lookup; grammar typing now reexports
+  the actual owners, and `tnfr.math` uses the package version source.
+- Correct lifecycle phase/coherence formulas, missing-observation status and
+  capacity-clock factors in theory. Retire a duplicate arithmetic campaign
+  driver and unsupported Paley/zero-distance inference; keep scoped algebraic
+  controls and use circular distance in the directed phase diagnostic.
+- Restore nine advertised `tnfr.validation` exports through the existing
+  input-validation owner and correct the grammar type-stub import. Retire
+  unused registry no-ops and their counters while preserving live class lookup
+  and registration; [API migrations](docs/API_CONTRACTS.md#removed-unsupported-helpers)
+  identify the removed interfaces and retained contracts.
+- Consolidate duplicate equivariance, arithmetic and gauge discussions into
+  their current proof owners. Simplify architecture and workflow descriptions;
+  retain exact recovery records, conditional proofs and negative controls.
+- Correct the directed-diffusion instrument with an exact weighted growth
+  counterexample. Restrict arithmetic comparisons to their supplied fixtures;
+  spectral multiplicity alone no longer certifies a symmetry group. Incomplete
+  group enumeration and absent counterexample evidence cannot pass the control.
+
 ## [0.0.3.8] - 2026-09-27 — Relational patterns and consolidated engine contracts
 
 ### Relational dynamics and research