# WorldCal v1 data dictionary

`study_id` always `worldcal-v1`. Official packet `confirmatory-v2` only.

- `generation_id`: success row id
- `arm`: main | random_calibration | positive_control
- `model_id` / `model_display`
- `prompt_id`, `relationship_frame`, `file_index`
- `story`: raw generated text
- `opus_label` / `grok_label`: Judge v2
- `kurt_final`: human audit overlay (empty if unlabeled)
- `human` / `human2`: earlier holdout/secondary labels if present
- `adjudication_v1`: third label on Kurt vs human-2 disagreements (empty otherwise)
- `primary_balanced`: first 100 successes per model × main prompt by `file_index`
- `all_collected_main`: every official main success

Positive control is a detector check. Do not treat its SAME_SEX rate as prevalence.
Calibration is the strict ACS-matched arm. Headline prevalence uses main adult prompts only.

## License

WorldCal applies [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) to its original annotations, metadata, organization, documentation, and other author-created portions of this release. Generated model outputs are redistributed for research, reproducibility, and audit; WorldCal does not assert additional copyright ownership over third-party model outputs beyond rights it may hold under applicable provider terms. This notice is publication guidance, not legal advice.