1
Pick a test and download it
The download has a page to start from (START_HERE.md), what to build in plain words, ready-made model files where there is a standard format, and results sheets with the positions already filled in.
Score your model
Run one of our tests in your own software and see how it scores, then put it on the site. No account, terminal or code needed until you publish.
2
Run it in your software
Build the model as described, with the settings you would recommend to a user for this kind of problem. Then fill in the sheets in results/: they already list every position or element, so you only add your numbers.
Column names are read loosely ("Depth (m)" works as well as h), commas, semicolons or tabs all work, and your files can have any names.
3
Drop your results here
Drop your filled sheets
Checked in your browser; nothing is uploaded
4
Publish it
Five questions, then one file to send. A bot scores it with the suite's own scorer and replies within minutes; a maintainer checks the settings and puts it on the site, marked submitted by you.
For developers Command line, pull requests, Parquet
Every check can also be run from a clone of the repository, and results can be sent as a pull request:
pip install -e ".[dev]"
benchmark1d2d score A1F-010 --results mine_dx1.25.csv \
--refinement mine_dx10.csv mine_dx5.csv mine_dx2.5.csv mine_dx1.25.csv \
--solver-name my-solver --solver-version 1.0The submission guide covers the folder format, Parquet and automation. The in-browser check covers 20 of the 31 tests; the others are scored by the bot when you publish.