Fish in a Pond — Visual Telemetry Benchmark

Scripted · no AI

Move the cursor over the ponds to steer a threat

DISCONTINUITY SNAP

Case 1: Programmed Reflex

Static Threshold Tree (no AI)

System 0

INITIALIZING

-- FPS

Tail

0.00 Hz

Spine

Tension

DISCONTINUITY SNAP

Case 2: Intuitive Preset Lottery

Weighted Preset Lottery + Lerp (no AI)

System 1

INITIALIZING

-- FPS

Tail

0.00 Hz

Spine

Tension

DISCONTINUITY SNAP

Case 3: Autonomous Organism

Blended Threat / Food Intent (no AI)

Dual-Process

INITIALIZING

-- FPS

Tail

0.00 Hz

Spine

Tension

Empirical Architecture Comparison Matrix

Feature / Characteristic	Case 1: System 0 (Reflex)	Case 2: System 1 (Preset Lottery)	Case 3: Dual-Process (Organism)
Decision Mechanism	Binary threshold tree (if/else)	Weighted lottery over preset dictionary	Continuous multi-axis blending of threat and food intent
Transition Fidelity	Instant snapping, jagged discontinuities	Smooth lerping between fixed designer presets	Dynamic continuous posturing & organic muscular tension
Hand + Food Conflict State	Rigid threat override (no nuance)	Anxious-Freeze mood selection	Ambivalent hesitation balancing hunger vs. survival
Main-Thread Performance	60 FPS (Zero CPU deformation)	60 FPS (GPU vertex deform)	60 FPS (Sandboxed multi-axis vector solver)

Fish Pond Benchmark Suite — Next.js 16 + Triforge + Strata CSS · all three controllers are scripted code, no AI model · Physical Safety Clamping: ACTIVE (all viewports)