SCALEML / README.md
chenyuwan666's picture
Link complete original Codex evidence and clarify export scope
890ae97 verified
|
Raw History Blame Contribute Delete
3.17 kB
metadata
language:
  - en
tags:
  - robotics
  - mujoco
  - simulation
  - agent-generated
configs:
  - config_name: chenyu-codex
    data_files:
      - split: runs
        path: chenyu/codex/manifest.jsonl

SCALEML

Public research artifacts from simulation-ready reconstruction experiments.

Available outputs

Chenyu's Codex MuJoCo generation sessions are under chenyu/codex. The run index records the track, item, run ID, model, budget, execution status and archive hashes. See the export description for details.

Each run has its own directory with output.tar.gz (the unchanged generated model, resources, previews and refinement attempts), run.json (selected public execution metadata) and files.json (file checksums and relative symlinks). Unpack an archive to obtain its output/ directory.

Generation completion or budget expiry is not a quality PASS. These uploaded sessions have not been independently scored in this export. Original local outputs and historical attempts are preserved.

The compact output exports exclude original input bundles, evaluator protocols, raw CLI session logs and credentials. The separate complete original Codex evidence export below includes the original staged inputs, logs and evaluator_protocol.json at the owner’s explicit request. Credentials and authentication caches are not included. This dataset publishes generated research artifacts; applicable rights and source terms for source-derived content still apply. No blanket license for third-party source media is implied.

Complete original Codex run evidence

The earlier compact Codex export remains unchanged. Complete originals for all 66 Mac Codex run folders are now published separately, including the byte-identical native campaign state, full run.json records, staged Agent inputs, submitted prompts, raw generation logs, recorded contracts, token usage, generated outputs and evaluator_protocol.json.

The original files were verified byte-for-byte (7,282 files and six relative symlinks). The state’s 294 dropped entries denote jobs handed off to Windows, not failed generation. Evaluator records are operator review evidence and must remain outside fresh generation Agent inputs. Independent quality scoring is not included.