Executable Layer
A-CSM is the bounded executable layer of the current public branch. It detects conversational contextual risk after interaction artifacts are available, within the public-safe release boundary.
A-CSM is not the whole research program and it is not a generic AI safety product page. It belongs downstream of CXC-7, CXOD-7, USCH, and USCI. Its role is to make part of the current public branch inspectable in executable form.
The public-safe core exists for reproducible inspection, external technical review, and discussion of a bounded release discipline. It does not claim clinical use, legal certification, or completed real-world validation.
The released core exposes a bounded, deterministic processing path.
A-CSM focuses on conversational contextual risk after interaction artifacts are available. It is designed to surface structured signals, not to make psychiatric or legal judgments.
A-CSM is not a runtime moderation gateway, not a refusal engine, not a medical diagnostic tool, and not a legal compliance certification layer.
Input. Structured conversation data in JSON or normalized Markdown transcript form.
Process. Deterministic de-identification, event handling, contextual inference, repeat-aware tracking, state derivation, and schema validation.
Output. Canonical reports containing status, stability, evidence traceability, digital fingerprint, and release-oriented GO or NO_GO results.
The public-safe core remains separate from confidential and governance-restricted layers.
