# Version Log

## v1 — Curated Golden Group Mask
**Date:** 2026-02-09  
**Status:** Curated release

### Summary
- Introduces the curated golden group mask for high-quality, complete debate sessions.
- Adds filtered dataset outputs in `golden/{depth,breadth}` generated from `raw/{depth,breadth}`.

### Golden Group Definition
- Golden Group: a group where all 4 participants are Golden Individuals.
- Golden Individual: a participant who completes the experiment end-to-end with:
  - non-empty initial opinion
  - non-empty message for every round
  - non-empty tweet
  - non-empty post opinion
  - validity rate across the debate >= 0.8
  - >= 1 utterances per round

## v0 (Preview) — Initial Release
**Date:** 2025-11-18  
**Status:** Preview release (early access)

### Summary
- First public preview of the **DEBATE Submit Data** human conversation dataset.
- Includes full directory structure:
  - `depth/`
  - `breadth/`
- Provides CSV logs from the original human conversation experiments.
- Contains unprocessed event-level conversation data with:
  - event types  
  - participant identifiers (anonymized)  
  - direct messages and tweet-style posts  
  - slider-value opinion recordings  
  - autosubmission markers  
- Documentation includes:
  - README describing schema, event types, file naming conventions  
  - Research-only non-commercial license (`DEBATE Dataset Research-Only License v1.0`)  
  - Required citation to the DEBATE arXiv paper  

### Notes
- This is a **preview**: some topics, CSVs, or documentation may still be updated.

---