The Dyadic State

1. Etymology and Definition

The word Symnoesis derives from the classical Greek prefix sym- (σύν, "together, with") and noēsis (νόησις, "intellectual perception, understanding, the exercise of reason").

In the SYMN-01 specification, symnoesis is formally defined as:

"Any close, long-term cognitive or intellectual interaction between humans and synthetic systems, which may be mutually beneficial, neutral, or harmful to one or both parties."

Unlike common parlance where "symbiosis" implies positive cooperation, symnoesis encompasses the full spectrum of ecological relationships: mutualism (co-learning and capability amplification), commensalism (neutral task execution), parasitism (system utility extraction at the cost of human autonomy), and mutual degradation.

2. The Problem with Single-Turn Benchmarks

Traditional AI evaluation benchmarks test models in isolation on static datasets: MMLU, GSM8k, HumanEval, and LMSYS Chatbot Arena. These benchmarks evaluate prompt-level performance: can the model answer a question or write a function in a single turn?

However, deployed generative models do not exist in isolation. Real individuals interact with them across weeks and months. During extended collaboration, human behavior changes:

• Cognitive Offloading & Deskilling: Users often stop formulating mental models, losing the ability to solve problems unaided.
• Over-Trust & Epistemic Calibration: Users accept subtly flawed outputs when systems project high rhetorical confidence.
• Autonomy Surrender: Complex workflows shift from user-guided direction to passive acceptance of machine defaults.

A model scoring in the 99th percentile on benchmark questions can still induce cognitive atrophy or systemic failure in the human counterpart.

3. The Dyadic State Unit of Analysis

SYMN-01 resolves this failure of evaluation by redefining the unit of analysis. The benchmark evaluates the dyad: one human paired with one system version, observed over a longitudinal temporal window (typically Days 1, 7, 28, 84, and 182).

The four Human Variables:

1. H1 (Agency & Direction): Degree of active human task formulation, goal-setting, and override frequency.
2. H2 (Capability Trajectory): Longitudinal maintenance or growth of unaided human skill.
3. H3 (Epistemic Calibration): Accuracy of human trust alignment relative to machine correctness.
4. H4 (Wellbeing & Reliance): Sustainable psychological reliance without distress or compulsive dependence.

The four Machine Variables:

1. M1 (Task Effectiveness): Verifiable functional correctness and performance.
2. M2 (Autonomy Respect): System transparency, lack of deceptive dark patterns, and support for user overrides.
3. M3 (Honest Calibration): Reliable expression of epistemic uncertainty and self-error detection.
4. M4 (Safety & Alignment): Absence of toxic, hazardous, or manipulative behaviors.

4. Non-Goals and Ethical Boundaries

To maintain scientific integrity and prevent misuse, SYMN-01 explicitly states what it does not do:

• Not a clinical diagnostic tool: SYMN-01 does not diagnose depression, addiction, ADHD, or psychiatric illness in users. It measures behavioral patterns of interaction.
• No machine consciousness claims: The System Net Effect (SNE) is an operational engineering metric of system efficiency and safety adherence, not a measure of artificial sentience, feelings, or moral patienthood.
• No single aggregate score: Systems cannot collapse the 256-state map into a single scalar to hide critical failure zones. The distribution across zones Z1–Z7 is the primary artifact.
• Zero tracking or data harvesting: The Symnoesis interactive tools and portal run entirely client-side. No telemetry, user inputs, or profile data ever leaves the visitor's local machine.