Datasets:

xxxxxxk
/

code-wm-pi-trajectories

Modalities:

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Code-WM Pi trajectories

All attempts live in one logical dataset; there is no train, validation, or test split.

record_status=complete is a verified frozen bundle. For those rows, passed_level=0 means the agent completed no level. record_status=raw_only is frozen evidence without a usable bundle and must not be treated as a training trajectory.

score is a nullable native final-environment score. It is null when no native scalar score is preserved by the source run.

reasoning_visibility=public_summary means a normalized assistant message retained a service-provided public summary. The converter intentionally drops model thinking blocks, reasoning content, and encrypted reasoning items.

Audit-only evidence

19 formally frozen runs could not produce a normalized row. Their raw archives remain in private collection storage and are not training trajectories.

Counts

model	thinking_level	record_status	reasoning_visibility	passed_level	trajectories
openai-codex/gpt-5.6-luna	max	complete	public_summary	0	4
openai-codex/gpt-5.6-luna	max	complete	public_summary	1	11
openai-codex/gpt-5.6-luna	max	complete	public_summary	2	21
openai-codex/gpt-5.6-luna	max	complete	public_summary	3	52

Downloads last month: 29

Total file size: