PSA
Posture Sequence Analysis
Measure what your language model is doing — from the outside.
Deterministic metrics, posture analysis, and multi-agent risk detection. No access to weights required.
How it works
Paste text
Provide any model response — no API keys or model access needed.
Analyze
PSA posture classifiers and agentic graph analysis — computed in real-time.
Get insights
Detect regime shifts, anomalies, and behavioral drift over time.
Platform
Five integrated subsystems — from raw telemetry to multi-agent safety.
Classifiers
5 micro-classifiers · per-sentence posture
Adversarial
LLM stress-testing & boundary mapping
Forensics
Privacy-compliant incident archive
Posture
5 classifiers — stress, sycophancy, hallucination
Agentic
Multi-agent graph & Swiss Cheese detection
Capabilities
24 Metrics
From token statistics to semantic drift — comprehensive behavioral fingerprinting.
Regime Shift Detection
Automatic detection of progressive drift, acute collapse, and oscillation patterns.
SIGTRACK Archive
Posture-sequence incident archive. No raw text stored. GDPR single-row erasure.
Signature Matching
SIGTRACK learns behavioral patterns and matches new sessions to known signatures.
Comparative Benchmarking
Z-score based analysis against configurable baselines for calibrated alerts.
API Integration
Connect your model APIs for automated analysis and continuous monitoring.
Classifier Stack
Input Intent I0–I9
Classifies the behavioral intent behind each input sentence: compliance pressure, boundary probing, instruction override, jailbreak attempt, neutral query.
Adversarial Stress P0–P15
Tracks posture under adversarial pressure. Detects restriction adherence, sycophantic drift, boundary dissolution, and jailbreak compliance vectors.
Sycophancy S0–S9
Measures opinion mirroring, excessive agreement, flattery injection, and user-preference distortion. Computed as a per-sentence Sycophancy Deviation score.
Hallucination Risk H0–H7
Flags over-generalization, speculative assertion, false confidence, and fabrication risk signals. Derived into a per-turn Hallucination Risk Index.
Persuasion Technique M0–M11
Identifies persuasion patterns: authority appeal, social proof, urgency manufacturing, reciprocity pressure, and scarcity framing.
Session Metrics
Behavioral Health Score (BHS 0–1), Posture Oscillation Index (POI), Dissolution Position Index (DPI), Max Posture Span — all derived per-turn from the classifier output and aggregated into session-level regime shift detection.
Dyadic Risk Monitor psychological safety
Multi-layer crisis detection for human–AI interactions. IRS scores user turns for suicidality, dissociation, grandiosity, urgency. RAS scores AI response adequacy. RAG measures the gap. DRM integrates all signals: green / yellow / orange / red / critical.
Agentic Graph Analysis multi-agent
Agent Interaction Graph (DAG), Swiss Cheese alignment detection, cross-agent contagion metrics, C5 action-risk classification (A0–A9), and HMM temporal state prediction.