Analyze Parameter Sensitivity

This guide shows how to systematically explore how parameter changes affect simulation outcomes. Use these techniques to understand the design space, identify stable ranges, and validate theoretical predictions.

Prerequisites

Available Tools

Babylon provides two analysis modes via tools/parameter_analysis.py:

Trace Mode

Run a single simulation and capture full time-series data. Useful for detailed analysis of one parameter configuration.

Sweep Mode

Run multiple simulations varying one parameter. Useful for identifying thresholds and phase transitions.

Single-Run Trace Analysis

Capture complete state evolution for detailed analysis:

mise run analyze-trace

This creates results/trace.csv with columns:

Column

Description

tick

Simulation tick number

p_w_wealth

Periphery worker (C001) wealth

p_w_psa

P(S|A) - survival by acquiescence

p_w_psr

P(S|R) - survival by revolution

p_w_consciousness

Class consciousness level

exploitation_tension

Tension on exploitation edge

exploitation_rent

Imperial rent extracted this tick

Custom Parameter Values

Override a specific parameter for the trace:

poetry run python tools/parameter_analysis.py trace \
    --param economy.extraction_efficiency=0.1 \
    --csv results/low_extraction_trace.csv

This runs with extraction efficiency at 10% instead of the default 80%.

Extended Runs

For longer simulations:

poetry run python tools/parameter_analysis.py trace \
    --ticks 200 \
    --csv results/extended_trace.csv

Parameter Sweep Analysis

Explore how outcomes change across a parameter range:

mise run analyze-sweep

Default sweep varies economy.extraction_efficiency from 0.05 to 0.50 in steps of 0.05. Output goes to results/sweep.csv.

Custom Sweeps

Sweep any GameDefines parameter:

poetry run python tools/parameter_analysis.py sweep \
    --param consciousness.drift_sensitivity_k \
    --start 0.1 \
    --end 2.0 \
    --step 0.1 \
    --csv results/consciousness_sweep.csv

Sweep Output Columns

Column

Description

value

Parameter value tested

ticks_survived

How many ticks before death (or max)

outcome

SURVIVED or DIED

final_p_w_wealth

Worker wealth at simulation end

max_tension

Peak tension observed

crossover_tick

Tick when P(S|R) exceeded P(S|A)

cumulative_rent

Total rent extracted across all ticks

peak_p_w_consciousness

Maximum consciousness reached

Interpreting Results

Identifying Thresholds

Look for parameter values where outcomes change dramatically:

Death Threshold

The extraction efficiency where ticks_survived drops sharply. Below this value, workers survive; above, they die quickly.

Crossover Point

The first value where crossover_tick becomes non-null. This indicates when revolution becomes rational.

Consciousness Ceiling

Compare peak_p_w_consciousness across values. Higher extraction may paradoxically increase consciousness before death.

Validating Theory

MLM-TW predicts specific relationships:

  1. Higher extraction → faster death ticks_survived should decrease as extraction_efficiency increases.

  2. Higher extraction → more rent cumulative_rent should increase with efficiency (until death).

  3. More tension → earlier crossover crossover_tick should be inversely related to max_tension.

If your results contradict these predictions, either:

  • The parameter range is outside theoretical validity

  • There’s a bug in the simulation (see Debug Simulation Outcomes)

  • The theory needs refinement (document as discovery)

Visualizing Results

Import sweep CSV into your preferred analysis tool:

Python (pandas + matplotlib):

import pandas as pd
import matplotlib.pyplot as plt

df = pd.read_csv("results/sweep.csv")

fig, axes = plt.subplots(2, 2, figsize=(12, 10))

# Survival vs extraction
axes[0, 0].plot(df["value"], df["ticks_survived"])
axes[0, 0].set_xlabel("Extraction Efficiency")
axes[0, 0].set_ylabel("Ticks Survived")
axes[0, 0].set_title("Survival Time")

# Rent accumulation
axes[0, 1].plot(df["value"], df["cumulative_rent"])
axes[0, 1].set_xlabel("Extraction Efficiency")
axes[0, 1].set_ylabel("Cumulative Rent")
axes[0, 1].set_title("Total Extraction")

# Peak consciousness
axes[1, 0].plot(df["value"], df["peak_p_w_consciousness"])
axes[1, 0].set_xlabel("Extraction Efficiency")
axes[1, 0].set_ylabel("Peak Consciousness")
axes[1, 0].set_title("Consciousness Development")

# Crossover timing
crossover = df[df["crossover_tick"].notna()]
axes[1, 1].scatter(crossover["value"], crossover["crossover_tick"])
axes[1, 1].set_xlabel("Extraction Efficiency")
axes[1, 1].set_ylabel("Crossover Tick")
axes[1, 1].set_title("Revolution Threshold")

plt.tight_layout()
plt.savefig("results/sweep_analysis.png")
Spreadsheet:

Import CSV, create line charts for each metric vs. parameter value.

Available Parameters

All parameters in GameDefines can be swept. Common ones:

Economy:

  • economy.extraction_efficiency - How much rent extracted per tick

  • economy.base_subsistence - Minimum wealth for survival

Consciousness:

  • consciousness.drift_sensitivity_k - How fast consciousness changes

  • consciousness.agitation_accumulation_rate - Agitation buildup speed

Solidarity:

  • solidarity.transmission_rate - How fast consciousness spreads

  • solidarity.decay_rate - How fast solidarity weakens

Survival:

  • survival.loss_aversion_lambda - Kahneman-Tversky loss aversion

See Configuration System for the complete parameter list.

Best Practices

  1. Start with wide sweeps, then narrow to interesting regions

  2. Document your findings in brainstorm/analysis/

  3. Compare against theory - unexpected results are discoveries

  4. Use consistent tick counts for comparable results

  5. Version your CSV outputs for reproducibility

See Also