PARS
The Petersen Accountable Reasoning Stack
PARS is a five-layer reasoning architecture for accountable AI in high-stakes operational decisions. It synthesizes five recently published reasoning architectures (EGuR, CLIO, DRN, AUQ, CGR) into one stack that produces visible reasoning, auditable state, operator steerability, and trust signals at every layer. PARS is the contribution this research program makes to the Orchestration stage of the Gen AI Value Architecture.
PARS at a Glance
Five layers, five architectures, five trust signals. Each layer produces a measurable output that boards, regulators, and operators can inspect.
The Five Layers
Strategy
EGuR
Match score against case memory.
Orchestration
CLIO
Trajectory shape and oscillation index.
Verification
DRN
Verifier verdict plus uncertainty value.
Memory
AUQ
Propagation history shape.
Termination
CGR
Certainty value at termination.
The Four Accountability Properties
Visible Reasoning
Every step of the reasoning process is observable and inspectable.
Auditable State
Complete audit trail of every decision and the state that produced it.
Steerable by Operator
Human operators can intervene, redirect, and override at any point.
Trust Signals at Every Layer
Quantified confidence measures at each stage of reasoning.
Context
PARS occupies the Orchestration stage of the Gen AI Value Architecture introduced in When AI Fails. The rest of the architecture remains operationally necessary; PARS replaces only the reasoning loop.
The active automotive diagnostic research program is the first operational test of PARS. Expert master mechanics validate ground truth. The first empirical results paper reports against the EGuR, CLIO, and DRN claims at the six-month window. Memory and Termination claims report at twelve months.
Research Program Timeline
High Stakes Reasoning
Executive paper released
Automotive Diagnostic Research
First operational test of PARS with expert mechanic validation
First Empirical Results Paper
Reports against EGuR, CLIO, and DRN claims
Clinical Transfer Paper
Extending PARS to clinical decision contexts
Integration Paper
Ahead of EU AI Act enforcement and NIST AI RMF maturity
Ready to discuss accountable AI for your organization?
Schedule a thirty-minute strategic briefing on PARS, OrbisFramework, and the automotive research in progress.