{"data":{"kind":"file","path":"README.md","version_id":"rt84397dq39f7vwtm6hwu8yk","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2279,"modified_at":"2026-06-02T20:34:58.077000","content_hash":"097127b0e2cb87d32641f19db46bc03eee14d387c8a886212ccecebaa1155462"},"entries":[],"content":"# rosalind-bio\n\nClassic [Rosalind](https://rosalind.info)-style bioinformatics problems as a\nsingle-turn environment for the Prime Intellect Environments Hub.\n\nEvery instance is generated programmatically, so the gold answer is computed\nexactly and the reward is fully deterministic — no external datasets, no LLM\njudge, no network access required.\n\n## Tasks\n\n| Type   | Problem                                          | Answer               |\n| ------ | ------------------------------------------------ | -------------------- |\n| `DNA`  | Count A, C, G, T in a DNA string                 | four integers        |\n| `RNA`  | Transcribe DNA → RNA (T → U)                      | RNA string           |\n| `REVC` | Reverse complement of a DNA string               | DNA string           |\n| `GC`   | GC-content of a DNA string (percentage)          | number (3 decimals)  |\n| `HAMM` | Hamming distance between two equal-length strings | integer             |\n| `PROT` | Translate an RNA string into a protein string    | protein string       |\n| `SUBS` | All 1-based positions of a motif in a string      | integers (space-sep) |\n\nInstances cycle through the task types. A `difficulty` knob controls sequence\nlengths, so the environment exposes a gradient of difficulty rather than a\nsingle trivial level.\n\n## Scoring\n\nThe rubric has two reward functions:\n\n- `correct_answer` (weight 1.0) — exact match on the extracted, whitespace-\n  normalized final answer.\n- `format_reward` (weight 0.1) — small bonus for ending the response with an\n  `ANSWER: <result>` line.\n\n## Arguments\n\n- `num_examples` (int, default 70) — number of generated instances.\n- `seed` (int, default 0) — RNG seed for reproducibility.\n- `difficulty` (str, default `\"mixed\"`) — one of `easy`, `medium`, `hard`, or\n  `mixed` (cycles through the three levels for a difficulty gradient).\n\n## Run a local eval\n\n```bash\n# install verifiers + the prime CLI, then:\nuv run vf-eval rosalind-bio -p prime -m deepseek/deepseek-v4-flash -n 14 -r 1   # smoke test\nprime eval rosalind-bio -m deepseek/deepseek-v4-flash -n 30                     # larger run\nprime eval rosalind-bio -a '{\"num_examples\": 90, \"seed\": 7, \"difficulty\": \"hard\"}'  # custom args\n```\n\n## Publish to the Hub\n\n```bash\nprime env push\n```\n","encoding":"utf-8","truncated":false,"total_bytes":2279},"status":null}