Feature Status Matrix
This page is the blunt version of the architecture story. The architecture pages are the construct; this page is the machine underneath. Every major Living Force feature lives in one of three states, and the label is whatever the disk actually says when you grep for it:
implemented— verified code or runtime wiring on the machine. Red pill. Plugged in.partial— meaningful implementation exists, but the doctrine still writes cheques the runtime can’t fully cash.documented only— the idea is in the docs; the cursor is blinking; nobody’s picked the pill yet.
The point is not to embarrass the architecture. The point is to stop the architecture and the current machine from being mistaken for the same thing. Every row below is a question: is this feature a row in a spreadsheet, or is it a file on the disk that survives rm -rf-level scrutiny. If the answer can’t be typed in one sentence, it’s a partial.
Status Summary
Section titled “Status Summary”implemented: 12partial: 0documented only: 0
Matrix
Section titled “Matrix”| Feature | Status | What was verified |
|---|---|---|
| Service Graph | implemented | service-graph.py exists, runtime manifests render into ~/.sanctum/services, and the checked-in audits prove graph edges such as qwen3-tts -> voice-agent (renamed from xtts-server 2026-04-25 after the Coqui→Qwen3 backend swap). |
| Immune System | implemented | The Rust watchdog, anomaly detector, remediation budget, incident-learning hook, and runtime graph now have an end-to-end harness that proves both self-heal and failed-escalation paths through the real watchdog API. |
| Code Forge | implemented | The real shared-skill scripts now have end-to-end proof for proposal creation, permission checks, validation, deploy, rollback, review, list, and audit logging. The live runtime also has a canonical agent-capabilities.yaml synced into place, so the permission model is no longer optional. |
| Tech Lookout | implemented | The real tech-lookout skill now has an end-to-end harness proving scan, briefing, and council dispatch from a dated report without depending on a hidden runtime. |
| Chaos Forge | implemented | Scenario scripts, preflight checks, verification tooling, and fire-drill orchestration all exist in the shared skills layer. |
| Evolution Loop | implemented | The real incident-learn.sh, perf-review.sh, and evolution-report.sh scripts now have an end-to-end harness that proves weekly synthesis, memory artifacts, and audit logging as one loop. |
| Genetic Health | implemented | genome-mcp now has a local proof path: the checked-in virtualenv runs all ten panels over synthetic 23andMe-style input, and the exported health profile on disk matches the same surface. |
| Centralized Calibration | implemented | The major generated artifacts now reconcile mechanically to their canonical inputs: workspace manifests, runtime manifests, agent capabilities, the instance.json cache, and the template-managed LaunchAgents all have explicit calibration checks instead of hopeful prose. |
| Navigator Sidecar | implemented | The workspace sidecar serves aggregate status, scrubs obvious secrets, and is covered by end-to-end tests. |
| Runtime Catalog | implemented | Live runtime manifests now render from instance.yaml plus the checked-in runtime catalog, and the audit enforces key dependency edges. |
| Kitchen Loop | implemented | The workspace now has a checked-in spec surface, canary suite, tribunal model, six-phase runner, and an end-to-end harness proving both a healthy cycle and a pause gate on canary escape. |
| Model Scout | implemented | model_scout.py proxy intelligence subsystem exists, natively hits OpenRouter and Google APIs, rates LLMs, commits findings to Memory Vault, and escalates adoption prompts directly to Qui-Gon via council-router. |
Main Gap
Section titled “Main Gap”The major Living Force feature set is now mechanically proven end to end. The remaining gaps are no longer feature claims; they are portability and packaging questions about how much of Sanctum should stay machine-local versus becoming productized.
~/Documents/Claude_Code~/.sanctum~/Projects/openclaw-skills
That split is survivable, but it means every architectural claim needs a scope. Without scope, the prose sounds more consolidated than the machine actually is.