---
configs:
- config_name: default
  data_files:
  - split: train
    path: index.csv
tags: [cooperbench, cooperdata, coop, mini-swe-agent, qwen, milestone-checkins]
size_categories: [n<1K]
---

## What this is

Cooperative two-agent coding dataset: 211 task pairs across 18 repos, generated with
`mini_swe_agent` on `Qwen/Qwen3.5-9B` in `coop` setting using a **milestone-checkins prompt variant** —
agents use periodic structured check-ins at predefined milestones to coordinate progress and
surface integration conflicts early. Patches are auto-merged after both submit.

**Coverage caveat:** Only 146 of 211 pairs were successfully evaluated (65 had eval errors).
The high agent Error rate (29.9%) is this run's primary data quality concern.

## At a glance

| Field | Value |
|---|---|
| Model | Qwen/Qwen3.5-9B |
| Agent | mini_swe_agent (milestone-checkins prompt) |
| Setting | coop |
| Repos | 18 |
| Pairs (results) | 211 |
| Pairs evaluated | 146 (65 had eval errors) |
| Both-pass | 2.1% (3/146) |
| Per-feature pass | 16.1% (47/292) |
| Merge clean rate | 65.1% (95/146) |
| Agent Error rate | 29.9% (126/422 slots) |
| Total tokens | ~89.1M (in+out, from traj files) |
| Owner | Arya Prabhudesai |

## How it was generated

```bash
cooperbench run --setting coop -a mini_swe_agent -c 30 qwen35-9b-milestone-checkins-coop
```

Model served via vLLM OpenAI-compatible endpoint (`openai/Qwen/Qwen3.5-9B`).
Milestone-checkins variant: agents perform structured coordination check-ins at key milestones
during their runs.

## File layout

- `index.csv` — one row per task pair; HF Dataset Viewer entry point
- `qwen35-9b-milestone-checkins-coop/coop/<repo>/<task_id>/<features>/` — raw per-pair artifacts:
  `result.json`, `eval.json` (where present), `agent1_traj.json`, `agent2_traj.json`,
  `agent{1,2}.patch`, `conversation.json`

`log_dir` column in `index.csv` points to the per-pair subdirectory.

## Schema highlights for mid-training

Filter on: `both_passed=true`, `model`, `agent_framework`.
`both_passed=null` means eval was not run for that pair (eval error).

`metadata` JSON carries per-agent statuses, steps, merge outcome, per-feature pass, and
`eval_missing` flag — use `json.loads(row["metadata"])` without following the pointer.

**Note:** `total_tokens` is 0 for this run — token counts are in `agent_full_traj.json`
under `messages[*].extra.response.usage` (~89.1M total in+out).

## Caveats

- Very high agent Error exit rate: 29.9% (126/422 slots) — likely caused by the milestone-checkins
  prompt variant triggering unexpected agent behaviour
- 65 pairs have no eval.json (eval errors): flask×12, jinja×9, typeguard×7, oauthlib×6,
  starlette×6, flake8×5, axios×4, sqlparse×4, trio×3, tweepy×3, pyparsing×3, indicatif×3
- Only 146/211 pairs evaluated — grade/pass-rate stats are partial
- 34.9% merge conflict rate (51/146 evaluated pairs)
- Token fields in `result.json` are 0; aggregate from `agent_full_traj.json` if needed

## Citation

```bibtex
@dataset{qwen35_9b_milestone_checkins_coop,
  title  = {qwen35-9b-milestone-checkins-coop},
  author = {Arya Prabhudesai},
  year   = {2026},
  url    = {https://huggingface.co/datasets/CooperBench/qwen35-9b-milestone-checkins-coop}
}
```