Master Kernel Library

The real, current, most-advanced version of every kernel type, in one place. Updated when something new solidifies, not before. Points at real files — doesn't reimplement them.

PurposeOne index instead of re-deriving "which version is current" from memory each session — the exact problem file-collisions and stale re-uploads keep causing.

StatusActive Development

Built fromDirect scan-back across every kernel-related file in the lab

Superseded by—

EvidenceLinks to real, individually-verified pages below.

→ Kernel Taxonomy — how all 15 registered kernels below are classified into 5 real families.

Governance

Current for execution-kernel fault-injection testing: governance_v2.py — TestingGovernance, real branching across ACCEPT/RETRY/QUARANTINE/REJECT, verified 4/4 acceptance tests. Current for state-estimation pipelines: GovernanceLayer (wired into the Pipeline Orchestrator) — a different governance for a different pipeline, not a superseded/replaced relationship. Which one a new pipeline should use is a real, still-open selection question — not yet resolved by a front gate.

Estimators (14-class suite)

Kalman, particle filter, ensemble, isolation forest, HMM, mixture of experts, Bayesian linear, plus the real diagnostic formulas in four_channel_formulas.py (schema_health_linear/quadratic, recovery_potential, channel_balance_entropy, cross_channel_coupling). Reconfigure for a new domain by swapping which formula feeds which channel — the estimator classes themselves don't change.

Orchestration

pipeline_front_gate.py (FrontGate, part registry) + multi_stage_pipeline.py (MultiStagePipeline, LINEAR/FOUR_CHANNEL stages). Reconfigure for a new domain by registering new formulas/estimators in the front gate's registries — the execution logic stays the same. Neither file independently executed on this site yet.

Decision layer

Current: the full Reasoning Chain — Observation Kernels, Research State Kernel, Decision Kernels. Closed gap: flagged by two independent external reviews as the highest-value missing piece — built and verified this pass.

Verification / tracking

Current: verification_registry.py (staleness tracking), document_metadata_box.py (Purpose/Status/Built-from/Depends-on/Evidence), exploratory_heuristic_42.py (tiered N-based confidence — referenced across this library, not yet uploaded to this site).

Real, un-coded ideas found on this scan-back — not yet in the library

Observation kernels beyond evidence-completeness — context saturation, protocol compliance, confidence distribution. Only one of four proposed observation kernels is built so far.
Maturity-scale tagging (Idea → Designed → Implemented → Verified → Validated → Production) — proposed, not yet an actual field on any page.
Planner fuzz harness's archetype-sampling pattern — proven once (1000 real trials, 100% correct handling per its own report), not yet generalized into a reusable fuzzing kernel other subsystems could call.
The activity-gating mechanism from multi_engine_hyperweave_playground.py — real and working, but wired into that one exploratory file only. Could genuinely apply to the front gate itself.

On document size

Splitting further only if a real section outgrows what's useful in one file — not preemptively. This index stays small on purpose; it points to real files rather than duplicating their content.