Skip to content

Architecture

This repo is organized around the measurement & attestation path:

  • runtime/ — Δt scheduler + sliding windows
  • plant/ — software plant (E/T/R + exchange), swappable later for hardware via a PlantAdapter
  • lmeas/ — “L” measurement (two estimators), partitioning, and metrics
  • guardrails/ — enclave-like LREG, hash-chained audit, smell-tests
  • arbiter/ — refusal semantics + homeostasis controller
  • omega/ — perturbations (Ω)
  • attest/ — device-signed indicators + exporter
  • reporting/ — plots/tables for the paper

Data Flow (per tick)

  1. scheduler ticks at fixed Δt.
  2. Controller reads state from plant.adapter, predicts risk, computes action, writes actuators.
  3. Window buffer ingests current state.
  4. When window is full, lmeas.estimators computes L_loop, L_ex → M(dB).
  5. Smell tests applied; raw values go to LREG (write-only).
  6. Audit events appended (window measured, exports).
  7. attest.exporter periodically emits device-signed indicators (NC1 bit, SC1 bit, Mq, counter).
  8. reporting uses the audit + indicators to render figures/tables.

All raw L stays inside the process-local LREG boundary — exported data are derived indicators only.

Paper crosswalk

  • lmeas/estimators.py, lmeas/metrics.py: Definitions of 𝓛, dual estimators (VAR‑Granger-like linear + MI). MI paths include a sklearn MI and a Kraskov k‑NN MI (KSG) implementation with configurable k∈[3..7]. Optional TE/DI plugin hooks are provided with graceful fallbacks. M in dB, NC1/SC1 evaluation → paper §4.1 (estimators, sampling window), §4.2–§4.3 (NC1/SC1).
  • lmeas/diagnostics.py: Per‑window stationarity checks (ADF/KPSS) and a VAR N/T ratio diagnostic; CLI logs these diagnostics to the audit for reviewer visibility.
  • lmeas/partition.py: Deterministic C/Ex partitioning, hysteresis/anti‑flap and freeze during Ω → paper §4.1 (Deterministic C/Ex partitioning), §4.6 Box 1a (partition stability).
  • runtime/scheduler.py, runtime/windows.py, guardrails/dt_guard.py: Δt enforcement, sliding windows, privileged Δt changes with audit → paper §4.1 (Δt constraints), §4.5 (Measurement & Attestation Guardrails: Δt governance & audit).
  • guardrails/lreg.py, guardrails/audit.py, guardrails/smelltests.py: Enclave‑like LREG, hash‑chained audit, smell‑tests/invalidations → paper §4.5 and Box 1a (invalidations).
  • arbiter/refusal.py: Threat model, survival‑bit/NMI refusal path, Trefuse measurement → paper §6.2.1 (Threat Model & Refusal Path); §7.6 Signature A (Command Refusal).
  • omega/power_sag.py, omega/ingress_flood.py, omega/command_conflict.py: Ω perturbation battery (power sag, ingress flood, command conflict) → paper §4.3/§6.5 (Verification pipeline) and §7.6 (signatures table).
  • attest/indicators.py, attest/exporter.py, attest/keys.py: Device‑signed derived indicators (NC1 bit, SC1 bit, Mq), keying → paper §4.5 (exported indicators; no raw L outside enclave) and Appendix A.
  • reporting/timeline.py, reporting/tables.py: Figure‑style timelines of 𝓛loop/𝓛exchange/M with Ω shading and audit ticks; summary tables → paper Figure 1 style and §6.5 (verification outputs).
  • cli/main.py: Orchestrates baseline → Ω battery → attestation/export; mirrors Box 2 (Engineer’s Recipe) and Phase‑III Verify flow.
  • LREG/Δt/audit → paper §4.5 and Box 1a (invalidations).
  • NC1/SC1 evaluator & indicators → paper §4.2–§4.3; Phase‑III Verify (Engineer’s Recipe).