A baseline Python implementation of the 1 Billion Row Challenge, used as a starting point for automated performance optimization with Hone.

The challenge: parse a text file with one billion rows of weather station measurements and compute the min, mean, and max temperature for each station.

hone-1brc/
├── data/
│   ├── weather_stations.txt   # Station list with mean temperatures (~45k stations)
│   └── measurements_*.txt     # Generated input files (not committed)
├── prepare.py                 # Generates the measurements input file
├── solution.py                # Computes min/mean/max per station
└── benchmark.py               # Times solution.py for use with Hone

pip install hone-ai

No dependencies beyond the Python standard library.

# Generate 1 billion rows (the full challenge)
python prepare.py 1000000000 data/measurements_1B.txt

# Generate a smaller dataset for testing
python prepare.py 1000000 data/measurements_1M.txt
python prepare.py 100000000 data/measurements_100M.txt

Output defaults to data/measurements.txt if no filename is given.

python solution.py data/measurements_1M.txt

Example output:

{Abha=-23.0/18.0/59.2, Abidjan=-16.2/26.0/67.4, ...}

Each line in the measurements file follows the format:

StationName;Temperature

Constraints:

• Up to 10,000 unique station names
• Station names: 1–100 bytes
• Temperatures: -99.9 to 99.9 (always one decimal place)

benchmark.py times solution.py end-to-end and prints the elapsed time in a format Hone can parse.

# defaults to data/measurements.txt
python benchmark.py

# explicit input file
python benchmark.py data/measurements_1M.txt
# Time Taken: 0.569

To run Hone against the baseline:

hone \
     --goal-file program.md \
     --bench "python /Users/rathla/workspace/hone-1brc/benchmark.py /Users/rathla/workspace/hone-1brc/data/measurements_1B.txt" \
     --files "solution.py" \
     --optimize lower \
     --score-pattern "Time Taken:\s*(\d+\.\d+)" \
     --budget 3.0 \
     --max-iter 50 \
     --model claude-haiku-4-5

The baseline solution is intentionally naive — no tricks, no concurrency, no memory mapping. It reads the file line by line, accumulates stats in a plain dict, and prints sorted results. This establishes the performance floor that Hone will attempt to optimize automatically.