ncfca_lib — Module Status

Consolidated Python package for the Non-Collapsible Four-Channel Architecture. A real audit caught 3 stale imports that had slipped through an earlier "done" claim — noted below, not hidden. Nothing here is asserted without having been actually run.

core/ — the architecture itself

• tensor.py — immutable, provenance-tracked (TaintedValue) channel tensor. The laundering vulnerability found and fixed this session is closed here.
• kernel.py — deterministic execution kernel. Byte-identical output across repeated runs, same seed.
• governance.py — TestingGovernance, real branching across ACCEPT/RETRY/QUARANTINE/REJECT. Had a stale flat-import (ncfca_execution_kernel_v4) that broke real package-level import — found and fixed this pass. Correction from later in this project: this is fault-injection governance for the execution kernel specifically, not general-purpose pipeline governance — see the Master Kernel Library for how this differs from GovernanceLayer, which governs the state-estimation pipeline.
• genesis.py, lattice.py — burn-in and state-machine layer. lattice.py is the corrected version (markdown-artifact syntax error fixed earlier).
• verification_registry.py — staleness-tracking for claims verified against an outdated invariant set.
• epistemic_protocols.py — executable versions of the 8 standing verification rules.
• ensemble_estimator.py — bootstrap-resampling fix applied. Also had a stale flat-import (base_estimator) found and fixed this pass.

estimators/ — the full 14-class suite

All 14 classes the package's own __init__.py was always designed around, assembled from files scattered across many upload rounds: BaseEstimator, ParticleFilterEstimator, ChannelCoherenceEstimator, ChannelDominanceEstimator, RecoveryPotentialEstimator, ChannelBalanceEstimator, BayesianLinearEstimator, IsolationForestEstimator, SimpleHMMEstimator, MixtureOfExpertsEstimator, ReferenceBaselineEstimator, ReferenceKalmanEstimator, ReferenceEnsembleEstimator, ReferenceParticleFilterEstimator. Verified: all 14 import together, one ran a real prediction end to end.

claims/ — the seven patent claims

Claims 2, 3, 6, 7 included. channel_nodes.py had a stale flat-import, fixed this pass. Still stale against the value-provenance invariant discovered during the Claim 1 laundering fix — real, open, not yet closed, tracked by verification_registry.py.

kernel_history/ — the real development sequence, kept, not overwritten

STATE_ESTIMATION_KERNEL_v1_0.py → v2.py → v8.py: a genuine, honest evolution (SimpleRegime/SimpleResult → Regime/EstimationResult → DiagnosticRegime/StateEstimationResult), confirmed by real class-name and structure comparison, not assumed from filenames.

STATE_ESTIMATION_KERNEL_v8_REAL_DEPS.py — a variant of v8 that expects real module dependencies (core_v2, detectors_tilt_detector) rather than the flat/fallback versions. Verified: imports cleanly now that pipeline_v2/ actually supplies those dependencies — likely the first time this specific variant has run against its true dependency chain in this project.

pipeline_v2/ — real dependencies for the v8_REAL_DEPS kernel

core_v2.py, detectors_tilt_detector.py, coordinator.py, base.py, metrics.py, pipeline.py, kalman.py. Exists specifically to satisfy kernel_history/STATE_ESTIMATION_KERNEL_v8_REAL_DEPS.py.

applications/ — domain-specific tools, not yet all individually run

TiltDetector.py, both CorrelationEvidenceCompilation.py and CORRELATION_EVIDENCE_COMPILATION.py (genuinely different files, checked directly — not a duplicate, both real), poker_decision_state_estimation.py, example_full_pipeline.py, real_data_backtest_example.py, SIMULATION_ENGINE_v1_0.py and v2.py, POKER_BOT_POPULATION_GENERATOR_v1_0.py, UPGRADE_TECHNIQUES_DETAILED_IMPLEMENTATION.py, lly_overreaction_walkthrough.py and _simple.py. All confirmed to compile; not all confirmed to run end to end yet.

validation/ + tests/ — the evidence

full_stack_test.py, isolation_baseline.py, replications.py, fault_injection.py, strong_alternative.py are the post-laundering-fix versions — deliberate substitutions over pre-fix uploads that would silently reintroduce the closed vulnerability.

tests/adversarial_test_suite.py — real, honest discrepancy: filename says 15 stress tests, actual count is 13 (test_01 through test_13). Worth knowing before this filename is cited anywhere as evidence of 15 tests.

Known gaps, stated plainly

1. Claims 2, 3, 6, 7 stale against the value-provenance invariant. Open.
2. Test 5's original script never located — reconstruction exists separately (TEST5_TUNED_FINAL_REPORT.md (pending source)), not yet folded into this package.
3. simulate_trading_session.py and the CSTEProbe/market layer — not included, depends on phase1_trader_state.py, which doesn't exist anywhere uploaded.
4. The real ncfca package itself is still missing entirely. Confirmed blocking at least two real files now (run_vector_experiments.py, all_estimators_demo.py, governance_demo.py). Its real internal structure is partially known from import statements even without having it: needs pipeline.pipeline.Pipeline and semantic.types.Hypothesis at minimum.
5. applications/ files are syntax-verified, not all run-verified yet — the next real audit pass should exercise each one directly, the same way estimators/ and pipeline_v2/ just were.

Running the validation suite

pip install -e .
cd ncfca_lib/validation && python replications.py
python tests/test_kernel_run.py