Skip to main content

Implementation status

This page is the current-state companion to the historical whitepaper. It is based on committed Artemis City dev @ 9d128f7b3acb12a1c6881ed962eff213285b1e58 and the recorded public/private boundary. Source-level evidence is not evidence of a deployed service.
No navigation entry is added for this reference page. Link to it directly when distinguishing current source behavior from the historical whitepaper.

Implemented in the committed source

  • Routing: the capability router is composite-led and can include persisted Hebbian and optional trust contributions. Governance state affects eligibility; this is not a pure Hebbian monopoly. See src/integration/hebbian_router.py.
  • Hebbian persistence: updates use default learning rate 0.1, per-outcome decay 0.99, and a weight floor of 0.01.
  • Sentinel alert: after a 50-sample warm window, a persisted rolling sign-change alert is recorded only when its rate is strictly greater than 0.4. It is observational: it does not reroute, down-rank, quarantine, gate, or change weights or trust.
  • Trust: trust policy is configurable; default numeric trust blend and floor are both 0. Passive daily trust decay is floored at the existing tier threshold and cannot, by itself, demote a tier. See src/integration/trust_interface.py.
  • ATP: ATP means Artemis Transmission Protocol. Parser completeness requires Mode, Context, and Action Type; TargetZone is advisory, not an enforced allow-list. See src/agents/atp/.
  • Recovery lifecycle: rollback requires confirmation and integrity validation, followed by bridge restoration. The memory-decay service runs a boot-time lifecycle cycle; no continuous all-memory deletion scheduler is claimed.

Simulation-only and historical material

The whitepaper’s benchmark comparisons, latency and cost figures, domain-monopoly claims, active Sentinel rerouting experiments, reconciliation outcomes, monitoring dashboards, and production-readiness assertions are historical architecture or simulation material. They are not current source or runtime metrics.

Planned or private work

Oracle Watcher and the public/private boundary place Oracle Watcher, L0–L6 tiers, capability tickets, and CNKD in private The Oracle work. That system is live in its private scope but remains planned/private from this public repository. The public historical “Sentinel” detector has a naming collision with Oracle Watcher and must not be treated as the same system.

Not claimed here

  • A public deployment, live service, or production benchmark.
  • Automatic standalone rollback, active Sentinel enforcement, or a continuous memory-deletion scheduler.
  • Public implementation of private Oracle Watcher behavior.
Last modified on August 1, 2026