Compliance and Local Safety
These notes define the final safety boundary beside the host. Pinky classifies observation sources, limits passive ingestion, isolates client-specific illusions, and may reject or debounce an approved operation when current local evidence makes it unsafe.
Local protection does not silently rewrite Brain's decision. It emits bounded evidence so authoritative state can be reconciled and repeated interference or incompatibility can be diagnosed.
Watching Without Changing
ADK can run alongside an existing plugin system in a passive way. It reads approved logs and data from the game to support health checks, analytics, and moderation. It does not change gameplay or inject into the game itself.
The information it collects is not as complete as what full integration would provide. Every result shows where it came from, how confident it is, and what parts are missing. Administrators can evaluate part of ADK without pretending that observation-only data provides precise entity truth, in-game UI, or intervention capabilities it cannot safely supply.
Stopping Unsafe Actions
Pinky holds trusted code that stops unsafe actions before they reach the main Brain system. It blocks dangerous interactions but does not let plugins run their own code.
These safety checks happen quickly and are logged for review. Only approved framework modules can be registered. Pinky interceptors do not execute plugin code. Brain and Pinky are separate safety boundaries. All interactions must stay within these limits.