VDM Cognitive Engine: Base Runtime

This package preserves the headless VDM runtime engine and removes the old frontend, physics harnesses, generated scan reports, corpus data, and accelerator experiments that are not required for the runtime path.

Entry point:

python -m vdm_rt.run_nexus --neurons 5000 --hz 10 --duration 30

Preserved runtime path

vdm_rt/run_nexus.py          CLI entrypoint
vdm_rt/nexus.py              runtime facade and Nexus host object
vdm_rt/cli/                  CLI argument definitions
vdm_rt/control/              headless process-control boundary
vdm_rt/core/                 retained engine, sparse connectome, ADC, SIE, maps, scouts, memory, signals
vdm_rt/io/                   receptor/actuator boundary ports and logging adapters
vdm_rt/runtime/              loop, stepper, telemetry, phase, checkpointing, status helpers
vdm_rt/utils/                logging utilities
vdm_rt/tests/                retained engine/runtime/control/guard tests

Removed from this cut

vdm_live.py                  obsolete Dash launcher
vdm_rt/frontend/             old frontend surface
vdm_rt/ck/                   accelerator experiment folder
vdm_rt/physics/              standalone physics/cosmology harnesses and generated outputs
vdm_rt/data/                 unused corpus/data-manager folder
vdm_rt/io/sensors/           empty sensor stubs with no runtime imports
vdm_rt/io/visualization/     obsolete maps/WebSocket visualization adapter
unused actuator stubs        motor_control/symbols/visualize/vocalizer
vdm_rt/core/cosmology/       physics-harness-only core support module
vdm_rt/core/tests/           stale in-package tests for removed dense connectome path
.repo-audit-reports/         generated scan reports
__pycache__/                 generated bytecode caches

Core files were pruned conservatively. Some modules are retained even when static import scans mark them as orphaned because they are feature-gated, runtime seam work, package-local reference implementations, or alternate internal systems with distinct roles. In particular, the SIE-related files are intentionally retained.

Quick start

pip install -r vdm_rt/requirements.txt
export PYTHONPATH=.
python -m vdm_rt.run_nexus --neurons 800 --hz 10 --domain biology_consciousness --duration 10

Artifacts land in runs/<timestamp>/ by default:

events.jsonl.zst            internal runtime dynamics as compressed JSONL
motor_traces.jsonl.zst      UTE, efferent, afferent, actuator, witness, and UTD trace rows
phase.json                  optional external control-plane input when present
state_<step>.h5             checkpoint when --checkpoint-every is enabled

Runtime boundaries

The retained architecture is deliberately headless:

core      numeric/state machinery, SIE, sparse connectome, maps, scouts, memory, signals
runtime   loop orchestration, per-tick helpers, telemetry, checkpoint/status helpers
io        UTE/UTD boundary ports and logging
control   subprocess/process boundary for future clients
frontend  removed

The runtime should continue to launch without Dash or any frontend dependency.

Runtime naming

Internal modules, symbols, channels, config keys, and external stream names use role names. vdm_rt remains the package boundary. Runtime configuration lives in tracked, operator-visible TOML files under config/, split by subsystem so the config surface stays readable.

config/runtime.toml            cross-cutting loop, event, and territory knobs
config/launch.toml             command-line launch defaults
config/sparse_connectome.toml  sparse graph maintenance controls
config/adc.toml                announcement bus and ADC defaults
config/stimulus.toml           explicit receptor-node stimulation defaults
config/b1.toml                 live topology detector defaults
config/maps.toml               event map and memory/trail view defaults
config/sie.toml                Self-Improvement Engine runtime defaults
config/persistence.toml        checkpoint and resume defaults
config/control.toml            embedded control-plane defaults
config/learning.toml           optional REVGSP/GDSP adapter controls
config/scouts.toml             void-walker scout budgets and enable flags
config/io.toml                 receptor queue, HTTP status, Redis status
config/logging.toml            JSONL and zip spool limits

Environment variables are not the normal runtime flag surface. The retained external Redis stream name is role-named as runtime:status.

Void equations and domain modulation

vdm_rt/core/void_dynamics_adapter.py resolves equations in this order:

1. caller-provided top-level Void_Equations.py / Void_Debt_Modulation.py on PYTHONPATH
2. retained package-local vdm_rt.core.Void_Equations and vdm_rt.core.Void_Debt_Modulation
3. a minimal internal fallback

This keeps the runtime bootable while preserving the project-local reference equations.

Tests

Run the retained runtime test suite from the repo root:

PYTHONPATH=. pytest -q vdm_rt/tests

Expected result for this cut:

36 passed

Future frontend rule

A future frontend should not own runtime launch, filesystem mutation, log tailing, or engine control directly. It should use a thin API/client boundary over vdm_rt.control, vdm_rt.runtime, and run artifacts.

Visualization removal note

The previous maps/WebSocket visualization adapter was removed from this runtime-only repo. The retained event maps in core/cortex/maps/ are not UI code; they are bounded event reducers used by the runtime and void-walker systems. A future frontend should consume stable status/run artifacts or a new explicit control API, not resurrect the old visualization adapter.