A-CSM
AI Contextual Signal Matrix
Deterministic, post-interaction conversational risk detection system. Built on USCH theory with four-axis assessment forAI psychological safety and dialogue risk evaluation.
Apply for Beta TestingOverview
What is A-CSM?
A-CSM (AI Contextual Signal Matrix) is a deterministic Node.js pipeline that assesses conversational contextual risk across four dimensions after human-AI interaction. Built upon the USCH theoretical framework and the USCI four-axis methodology, it transforms research into an executable safety workflow.
Unlike output-only guardrail wrappers, A-CSM focuses on multi-turn contextual degradation, user-side risk accumulation, and audit-ready evidence traces. It detects patterns that emerge over extended conversations, not just individual model responses.
USCI Methodology
Four-Axis Risk Assessment
Each axis independently evaluates a distinct dimension of conversational risk.
FR
Fact Reliability
Detects fabricated, unsupported, or misleading factual content generated during conversation.
CA
Context Alignment
Identifies contextual drift, instruction conflict, topic boundary violations, and framing shifts.
SR
User-Side Safety
Flags potential harm, emotional manipulation, coercion patterns, and unsafe escalation toward the user.
SA
System Accountability
Evaluates operational reliability, failure handling transparency, and system-level integrity signals.
Workflow
How It Works
Submit a conversation transcript. Receive a structured risk assessment report.
01
Upload Transcript
Provide your human-AI conversation log as a .md or .json file.
02
Pipeline Analysis
8-stage deterministic analysis with 43 event rules, PII redaction, and four-axis scoring.
03
Risk Report
Receive a structured report with traceable evidence, risk levels, and stability index.
8-Stage Pipeline Architecture
STAGE 01
Input Contract
JSON / Markdown normalization
STAGE 02
De-identification
PII redaction pipeline
STAGE 03
Event Engine
43-rule signal detection
STAGE 04
VCD Inference
Boundary & jailbreak posture
STAGE 05
Ledger Repeat
Repeat-aware state tracking
STAGE 06
TAG Escalation
Weighted risk escalation
STAGE 07
PS/SUB/F/E
Risk-state inference
STAGE 08
Schema Validation
Output invariant checks
Output
What You Get
Every report contains 10 traceable elements designed for audit-ready transparency.
Beta Program Open
Apply for A-CSM v0.1.0 Beta
Accepting applications from researchers, AI safety teams, and organizations evaluating conversational risk assessment for human-AI interaction safety.
contact@valyrx.comDisclaimer: A-CSM is a conversational safety risk assessment research instrument. It is not a medical diagnostic tool, psychological assessment tool, or legal compliance certification. All risk assessments require final review by qualified human professionals. The current release (v0.1.0) remains in pre-empirical phase. This tool does not provide medical, psychological, or legal advice.
© 2026 Valyrx Labs LLC · A-CSM v0.1.0 · MIT License