Project 06 — End-to-End Research Project

Paired with: Lecture 11 + Lecture 12
Starter code: projects/project-06/starter/
Solution: projects/project-06/solution/

---

What You'll Build

A complete end-to-end automated research pipeline: choose your own optimization target, run an overnight simulation (50 iterations), debug any issues, run a security audit, and ship the result.

Learning Objectives

• Design your own research.md from a blank slate
• Run a simulated overnight session (50 iterations with max_iterations: 50)
• Chain all 10 commands in a single cohesive workflow
• Write a final_report.md and use it to scaffold the next research session

Your Mission

Choose ONE of the following optimization targets (or propose your own):

Option	Target	Metric
A	Optimize a text compression algorithm	compression_ratio (maximize)
B	Improve a recommendation system's precision	precision_at_10 (maximize)
C	Reduce a graph traversal algorithm's memory	peak_mb (minimize)
D	Your own domain (requires writing evaluator)	Your choice

The Full Pipeline

# Phase 1: Plan
/autoresearch:plan

# Phase 2: Research (simulate overnight — 50 iterations)
/autoresearch
Iterations: 50

# Phase 3: Debug (if any issues arose during research)
/autoresearch:debug

# Phase 4: Fix
/autoresearch:fix

# Phase 5: Security (if the artifact is an API or service)
/autoresearch:security

# Phase 6: Ship
/autoresearch:ship --type research

Deliverables

By the end of this project, you should have:

1. research.md — full experiment history (50 rows)
2. research_log.md — detailed notes on key iterations
3. final_report.md — best result + recommendations for next session
4. autoresearch-results.tsv — machine-readable results
5. progress.png — convergence plot

Reflection Questions

After completing the pipeline, write a short reflection.md answering:

1. How many iterations did it take to reach the target? Did you need to pivot?
2. What was the most surprising discovery in the research history?
3. If you ran another 50 iterations tonight, what would you try?
4. What would you do differently in the research.md setup?

Overnight Simulation Script

To simulate a real overnight run without actually waiting overnight:

# Run 50 iterations in tmux
tmux new-session -d -s overnight
tmux send-keys -t overnight \
  "claude -p '/autoresearch Iterations: 50'" Enter

# Monitor progress from another terminal
watch -n 60 "tail -5 autoresearch-results.tsv && echo '---' && cat research.md | grep 'Best:'"