Project 01 — Your First Research Loop

Paired with: Lecture 01 + Lecture 02
Starter code: projects/project-01/starter/
Solution: projects/project-01/solution/

---

What You'll Build

Run the canonical autoresearch example: optimize a Python sort function from ~2.4 seconds to under 0.5 seconds on 1 million integers. You'll use /autoresearch:plan to scaffold the project and /autoresearch to run the loop.

Learning Objectives

By completing this project, you will:

• Convert a vague goal ("make it faster") into a mechanical metric (median_time_s < 0.5)
• Write an evaluator that outputs {"pass": bool, "score": float}
• Run the 5-stage loop and read the resulting research.md history
• Understand what gets committed and why

Starting Point

projects/project-01/starter/
├── sort.py          ← the slow implementation (recursive quicksort)
├── test_sort.py     ← correctness tests (never modify)
└── task-prompt.md   ← your mission

sort.py contains a recursive quicksort on a list of 1 million integers. It takes ~2.4 seconds. Your goal: get it under 0.5 seconds without breaking the tests.

Step 1 — Run the Plan Wizard

cd projects/project-01/starter/
/autoresearch:plan

Answer the wizard questions:

• Goal: Reduce sort.py execution time to under 0.5s on 1M integers
• Metric: median_time_s, direction: minimize, target: < 0.5
• Noisy? Yes — benchmarks vary. Set noise_runs: 3
• Scope: sort.py only. Forbidden: test_sort.py
• Guard: python -m pytest test_sort.py
• Max iterations: 20

The wizard generates benchmark.py and research.md.

Step 2 — Run the Loop

/autoresearch

Watch the loop run. After 3–5 iterations you should see:

| 3 | radix sort base 256     | 0.871 | keep    |
| 4 | radix sort base 65536   | 0.573 | keep    |
| 5 | micro-optimized radix   | 0.498 | keep ✓  | ← target met

Step 3 — Read the Results

Open research.md and research_log.md. Answer these questions:

1. How many iterations did it take to hit the target?
2. Which experiments were discarded? Why?
3. What does the git log look like? (git log --oneline)

Expected Outcome

Final best: median_time_s = 0.498
Target: < 0.5 ✓
Iterations used: 5 of 20

Verification

python benchmark.py
# Expected: {"pass": true, "score": 0.498}

python -m pytest test_sort.py
# Expected: all tests pass

Hints

• If the loop is running slowly, it's because noise_runs: 3 runs the benchmark 3 times per iteration. This is correct behavior — benchmarks are noisy.
• If you see GUARD FAILED, the sort is producing wrong results. Check which change broke correctness.
• The solution directory contains a completed research.md with a full 5-iteration history you can compare against.