NS26-0002 · Diagnostic belief workspace

When a sensor keeps giving the same unexpected reading, should an agent repair a component, check the sensor, or ask for a different measurement? This blueprint designs a non-LLM agent for small simulated circuits. It keeps competing explanations for circuit faults and sensor calibration, predicts what each possible test would reveal, and can stop with an inconclusive diagnosis.

Workstation Rebellion chose Stanley Kubrick’s 1968 film 2001: A Space Odyssey as the conceptual framework. The research author drew on film catalogues and criticism to explore a specific idea: a capable system can still receive faulty evidence from its own instruments. The source register documents the materials consulted and their limits. Sources and reading scope

Section 8 shows one operating cycle. Section 10 follows a simulated case step by step. The handbook, evaluation plan, implementation plan and resource budget describe the system and how to test it.

Two original CPU studies examined diagnosis and sensor interpretation. A rule requiring two independent confirmations missed its stated error-reduction target. A model that knew the sensor law slightly outperformed the neural estimator on balanced log loss. Both outcomes informed the design. The public repository includes the protocols, programs, generated examples and measured results. Experiments · Results

On 11 November 2026 Workstation Rebellion adopted revision c1 as a bounded research specification after independent review. End-to-end circuit restoration, benefits from episodic learning and durability after a crash still require measurement. Review record

The lab is reviewing reuse terms for this edition. Its Git author and committer are Workstation Rebellion. Edition and credits · Project page