Experiment Lab

The Experiment Lab (services/experiment_lab.py, 4,209 lines) is Frank's autonomous research facility — 7 simulation stations that let him run real experiments without leaving his chair.

7 Experiment Stations

1. Physics Station

8 simulation types with classical mechanics:

Example: Frank hypothesizes that a 45° angle maximizes projectile range. He runs the experiment. The physics station simulates the trajectory. The hypothesis engine records: confirmed (in vacuum) or refuted (with air resistance, it's ~43°).

2. Quantum Lab

6 simulation types — actual quantum state vector math:

• Qubit Simulation: State vectors, gate sequences (H, X, Y, Z, T, S, CNOT), measurement probabilities
• Variational Quantum Circuits: VQC for logic gate approximation
• Quantum Effects: Tunneling (WKB approximation), double-slit interference, harmonic oscillator
• Entanglement: Bell states, CHSH inequality tests, quantum teleportation
• QUBO: Quadratic Unconstrained Binary Optimization — brute-force + simulated annealing
• World Model: Free energy minimization, Bayesian belief updates

Uses numpy for matrix operations, sympy for symbolic math. Not a toy — these are real quantum mechanics calculations.

3. Astronomy Orrery

N-body gravitational simulation with leapfrog integration.

5 presets: inner solar system, Earth-Moon, binary star, three-body problem, Jupiter system. Tracks energy conservation (ΔE/E%), orbit radii per body. You can ask Frank "what would happen if Earth were 10% closer to the Sun?" and get a numerically simulated answer.

4. Game of Life Sandbox

Conway's Game of Life with 8 preset patterns: glider, blinker, block, R-pentomino, acorn, pulsar, loafer, methuselah. Configurable grid size and step count. Frank uses this to test emergence hypotheses — simple rules producing complex behavior.

5. Electronics Station

Circuit analysis with real component math:

• Voltage Divider: Resistor voltage distribution
• RC Circuit: Charging/discharging curves, time constants
• RL Circuit: Inductor response, L/R time constants
• RLC Resonance: Resonant frequency, Q factor
• Bode Plot: Frequency response — magnitude and phase

6. Math Station

Symbolic and numerical computation via sympy:

• Polynomial root finding (quadratic through quartic)
• Symbolic differentiation
• Numerical integration (trapezoid, Simpson's rule)
• Matrix eigenvalue decomposition
• Linear system solving (Gaussian elimination)
• Fibonacci sequence generation with growth rate analysis

7. GAN Lab

Adversarial training simulations with actual PyTorch neural networks (~80-200 params each):

• Digit Synthesis: Generator vs. Discriminator (MNIST-style)
• Point Cloud: 2D generative modeling
• Mode Collapse: Multi-modal distribution learning analysis
• Balance: Adversarial equilibrium metrics

Budget & Safety

How It Works End-to-End

1. The Hypothesis Engine creates a hypothesis: "Friction coefficient > 0.5 prevents sliding on a 30° incline"
2. It calls request_experiment() with station=physics, type=incline, params={angle: 30, friction: 0.5}
3. The lab checks the daily budget (atomic increment in experiment_budget table)
4. The physics station runs the simulation — pure Python + numpy
5. The result is stored in experiment_lab.db with station, params, result JSON, and narration
6. The hypothesis engine interprets the result: confirmed/refuted/inconclusive
7. The hypothesis is updated with the experimental evidence

All of this happens autonomously while Frank is idle. No user input required.