thanks for checking out the bounty program for the Prime Intellect Environments Hub!

TLDR: you earn a bounty by successfully merging a PR from the spreadsheet. to get your PR merged.

the key criteria we’re looking for are:

TLDR:

• high-quality code:
  • fully functional — installing from the hub + running vf-eval should just work without any additional configuration
    • Exceptions: you can expect users to have set environment variables for API key names (but NOT for data paths or other configurable variables)
  • up-to-date with verifiers>=0.1.3.post0 + documented
  • follows ruff style
    • run ruff check --fix . before PRing / install the pre-commit hook
• self-contained code:
  • all data-loading and resource-configuring logic should run within load_environment.
  • in almost all cases, you should NOT manually save a preprocessed version of your dataset on huggingface. use the original source files, process on the fly.
• validated code:
  • you should run vf-eval -s to save sample outputs, inspect these with vf-tui, and include them in your PR
• let us know when you’re ready for review (if you haven’t already) by setting the PR to Open
  • you do not need to tag or mention us. we will review as time permits.

To understand best practices for environment development, please read the verifiers docs and see other merged examples.

If your PR was previously Open and is now set to Draft, this is us flagging that you have some TODOs remaining.

code quality guidelines

we shouldn’t have to tell you this, but PRs cannot not be “vibe-coded slop”. by all means, use AI tooling to speed up implementation and debug, but do not rely on it for making consequential design choices without checking them. models are prone to silliness for these kinds of tasks, and this can ruin the integrity of an eval. we ask that you take responsibility for understanding the original source in depth, and are prepared to answer questions about anything that looks funny in your code (for which “idk Claude did it” is not an acceptable answer).

• if we judge your submission to be clearly a bad-faith attempt at submitting a correct + high-quality implementation, we reserve the right to close your PR and release the lock on your task.
• if you have multiple bounties in progress, this can extend to your others as well.