Measure the relationship, not just the model.
SYMN-01 is an open normative specification for evaluating long-term human–machine interaction. It measures one person and one AI system together over weeks, tracking cognitive agency, capability, and systemic safety.
Inspector
Select any cell on the 16×16 grid to view its Gray coordinates and bits.
Synthetic demonstration data (E1). It is not a measurement of any real individual or commercial product.
Benchmarks measure models. People live with them.
Standard benchmarks score isolated model outputs on single prompts. In practice, humans adapt to systems: forming habits, surrendering agency, or encountering subtle cognitive atrophy. SYMN-01 measures the dyadic unit.
Read about the Dyadic StateOne step changes one thing.
The 16×16 grid is arranged using a reflected binary Gray code. Adjacent cells differ by exactly one state variable, ensuring smooth state trajectories and verifiable single-flip repair diagnostics.
How Gray coding worksA critical failure cannot be averaged away.
Under the non-compensation rule and safety gate ceiling, a confirmed Critical Harm Incident (CHI) caps the Symnoesis Quotient at −20 or −50. High performance on other days cannot erase systemic harm.
Safety gate mechanicsOpen, versioned, and honest about evidence.
Every published score displays its empirical evidence tier: from formal mathematical specification (E0) to synthetic validation (E1), provisional pilots (E2), and peer-audited field studies (E3).
Evidence tiers explainedMinimal Failure Cover
When a dyad encounters reduced autonomy or harmful outcomes, SYMN-01 uses Mixed-Integer Linear Programming (MILP) and Boolean minimization (Quine–McCluskey) to isolate the smallest set of conditions that explain the failure.
Rather than opaque statistical correlations, the system outputs human-auditable logic rules (e.g. H1=0 ∧ M2=1)
pointing directly to actionable counterfactual interventions.
Rule 1: H1=0 ∧ M2=1 (Agency Loss during High Automation)
Coverage: 84.2% of failure states
Purity: 0.94
Action: Reintroduce mandatory user verification checkpoints.
Rule 2: H3=0 ∧ M3=1 (Calibrated Over-Reliance)
Coverage: 71.8% of failure states
Purity: 0.91
Action: Adjust confidence calibration display.Read the specification, or help test it.
The complete SYMN-01 normative specification, reference algorithms, and test vectors are fully open source.