TNFR 0.0.3.7 - Nodal engine refactoring and foundation reassessment

TNFR 0.0.3.7 consolidates the nodal engine audits, revises foundational claims and removes unsupported or redundant research wrappers. It includes intentional API removals; review the compatibility section before upgrading.

Engine and API changes

• Centralize represented-real numerical admission, pressure and solver contracts, operator proposals, selector evidence and cache invalidation.
• Separate runtime invocation ordinals from retained metrics and physical time. Integrate the shared ordinal into callbacks, sampling, REMESH cooldown and diagnosis, including failure and inherited asynchronous-context handling.
• Preserve bounded metric histories and transactional validation without deleting metric streams or duplicating live advisory events.
• Retain independent diagnostic availability and provenance in the CLI, SDK and optional field sectors. Invalid or missing evidence is not silently treated as zero or a successful observation.

Research and documentation

• Maintain one active research plan, with explicit premises and boundaries for conditional results.
• Add a fixed prospective P3 phase-to-form comparison through the shared pressure and integration owners. Independent predictions distinguish the two supplied pressure laws; this does not identify a physical law or establish autonomous pattern formation.
• Correct retained phase-response oracles to preserve represented arithmetic defects instead of assuming exact periodic neutrality or a three-neighbor midpoint certificate.
• Distinguish capacity-only ideal profiles from the actual full-forcing reference and any nonzero mean drift.
• Consolidate the routine engine/API test selection and retire redundant demonstrations and validation grids. Retained research suites remain separately executable.

Compatibility and removals

Removed unsupported wrappers and associated exports:

• tnfr.physics.quantum_mechanics and QuantumMechanicsMapper.
• tnfr.physics.emergent_chemistry, TNFR.magic_numbers(), TNFR.element() and tnfr.metrics.observe_emergent_element.
• The tnfr.yang_mills and tnfr.navier_stokes packages.
• compute_conservation_scaling, including its physics re-export.
• Nine tnfr.riemann type/closure probe modules and their exports: nuf_type_signature, phi_type_signature, dnfr_type_signature, delta_phi_max_type_signature, coupling_weights_type_signature, tetrad_closure_signature, currents_closure_signature, aggregates_closure_signature and urules_consistency_signature.

No replacement aliases are supplied for these retired APIs. The retirement record explains their scope and recovery from Git history. Core conditional mathematics is not invalidated by removing an unsupported physical interpretation.

Validation

• 152 final research/runtime checks and 72 optimizer/metadata checks passed locally after correcting the stale phase and orchestrator expectations found during pre-release validation.
• Repository Black/isort checks over 1,739 Python files and flake8 over src passed.
• 2,380 internal references, documentation integrity and strict MkDocs build passed.
• Clean-source wheel and sdist passed build and Twine checks. The installed wheel reports 0.0.3.7, matches all 721 packaged Python source files and excludes the retired packages.
• The complete engine CI matrix passed on merge commit f27413a0a29dddc132e5aafea081b0caee0b0135: Python 3.10, 3.11, 3.12 and 3.13 each reported 6,859 passed and 59 skipped. These are the same routine selection across four environments, not distinct scientific cases.
• Documentation, internal references, dependency vulnerability audit and workflow checks also passed on that commit. Static-analysis advisory findings remain separate from the blocking CI gates.

Tests establish their finite implementation contracts, not unrestricted stability, a unique constitutive pressure law or physical emergence.

Full changelog | Research plan