Agentsociety Synthesize

Parameters

Two Modes

1. Auto-Discovery Mode (Recommended)

Run without arguments to automatically discover and synthesize all hypotheses and experiments:

$PYTHON_PATH scripts/synthesize.py

This will:

1. Discover all hypothesis_* directories
2. For each hypothesis, discover all experiment_* directories
3. Run analysis on each experiment (if not already analyzed)
4. Generate cross-hypothesis synthesis report

2. Targeted Mode

Specify particular hypotheses or experiments:

# Synthesize only hypotheses 1 and 2
$PYTHON_PATH scripts/synthesize.py --hypothesis-ids 1 2

# Synthesize only experiments 1 and 2 under hypothesis 1
$PYTHON_PATH scripts/synthesize.py --hypothesis-ids 1 --experiment-ids 1 2

What It Does

1. Per-Experiment Analysis: Analyzes each experiment (reuses existing analysis if available)
2. Hypothesis Summaries: Aggregates results per hypothesis
3. Cross-Hypothesis Synthesis: Compares and synthesizes across hypotheses
4. Report Generation: Creates bilingual reports (Chinese and English)

Output Files

synthesis/
├── synthesis_report_{timestamp}.md      # Markdown synthesis report
├── synthesis_report_{timestamp}.html    # HTML synthesis report
└── assets/
    └── synthesis_comparison.png         # Comparison chart

Content Sections

1. Synthesis Strategy - How the synthesis was approached
2. Cross-Hypothesis Analysis - Comparative narrative
3. Comparative Insights - Key differences and similarities
4. Unified Conclusions - Overall findings
5. Recommendations - Suggestions for future work
6. Best Hypothesis - If applicable, the best performing hypothesis

Comparison with agentsociety-analysis

Prerequisites

• At least one experiment should have been run
• For best results, run agentsociety-analysis on each experiment first

Documentation Sync

After synthesis completes, update:

1. TOPIC.md - Add synthesis summary to research progress
2. HYPOTHESIS.md - Add synthesis link and conclusions
3. EXPERIMENT.md - Add synthesis reference