---
license: apache-2.0
tags:
- spare
- generated-environments
- self-play
---

# qwen3-8B-0627-plateau-6skill-thinking-proposer — generated environments

Environments generated by the SPARE proposer during training run
`sh04swu4` (qwen3-8B-0627-plateau-6skill-thinking-proposer), recovered from the spare-viz durable cache.
The run's scratch directory no longer exists; this dataset is the surviving copy.

| | |
|---|---|
| Games | 1411 |
| Steps covered | 73 (step 0–149) |
| With recovered skill | 1312 |
| With hint | 0 |
| Actor / proposer model | `/scratch/spare-workspace/Qwen3-8B` |
| WandB segments | rsic4mcb, sh04swu4 |

## Layout

    manifest.json                                  authoritative games list
    games/gen_<NNNN>/game_<NNNNN>_<NNN>_<slug>.py  one environment per file

`generation` numbers are dense over the training steps actually captured; the true
training step is the `step` field. Each game exposes the standard SPARE contract
(`reset(seed=None)`, `step(action) -> (obs, reward, terminated, truncated, info)`).

## Loading

Load games through the project loader, not a bare `import` — it injects the common
stdlib names and the `ToolUseBaseEnv` / `TerminalBaseEnv` base classes that generated
games subclass without importing:

```python
from spare.core.envs.synthetic_game_env import make_synthetic_env
env = make_synthetic_env("games/gen_0000/game_00000_000_api_orchestration.py")
obs, info = env.reset(seed=0)
obs, reward, terminated, truncated, info = env.step("...")
```

## Caveats

- **Partial step coverage.** The viz extractor pulls newest-first with a call budget,
  so a run's captured steps are a subset of the steps it trained.
- **No joined rewards.** Weave payloads for these runs predate the reward join; mean_reward/solve_rate are null where reward_joined is false.
- `skill` / `difficulty` are parsed from the proposer prompt, not from a stored label.

Rendered in the env gallery via `SPARE_VIZ_ENV_DATASETS=<rid>=<this dataset>`.