AI Contextual Signal Matrix (A-CSM).
Technical framework for detecting, classifying, and monitoring cognitive drift signals in AI conversations. Builds on USCP findings to operationalize detection into signal taxonomies, a 43-event NLP codebook, and four-axis risk assessment.
/
What It Is
(01)
A detection engine for what the model cannot see about itself.
A-CSM operationalizes USCP research findings into a detection system. Where USCP names the phenomena, A-CSM provides the machinery: signal taxonomies derived from the USCP codebook, detection algorithms for identifying drift patterns in conversation transcripts, and four-axis risk assessment for evaluating contextual stability.
The framework addresses a specific gap: model-level safety tools measure what the model outputs. A-CSM measures what the user accumulates.
/
Technical Foundation
(02)
CXC-7, CXOD-7, and the PRISM-NLP detection pipeline.
Builds on three foundational papers: CXC-7 (seven core dimensions of conversational context), CXOD-7 (contextual offense-defense framework), and USCI (user-side contextual interaction assessment — four axes: Fact Reliability, Context Alignment, User Safety, System Usability).
The PRISM-NLP pipeline processes conversation transcripts against a 43-event codebook to identify drift-relevant signals. The CogLens module tracks six-stage cognitive change across conversation history. Detection outputs feed into USCI scoring: Normal, Deviation, and Alert classification.
/
Applications
(03)
Research monitoring and conversation analysis. Not clinical use.
Designed for research and monitoring contexts, not clinical application. Current targets: longitudinal interaction monitoring for researchers, single-session risk assessment for conversation analysis, and corpus validation tooling for USCP research.
The system is in active development. Published technical reports document current methodology, signal taxonomies, and known limitations. Independent replication and critique are welcomed.