Standardize the daily-file naming convention with TimeSeventeen/Polymarket-v2 (which uses YYYY_MM_DD.parquet). Previously this dataset's daily_aligned/ and daily_aligned_multi/ used YYYY-MM-DD.parquet (dash), the only inconsistency between the two datasets' otherwise-aligned schema/layout.
- daily_aligned/: 1248 files renamed (dash -> underscore)
- daily_aligned_multi/: 857 files renamed (dash -> underscore)
- Content unchanged (LFS-pointer-only rename via CommitOperationCopy+Delete, no blob re-upload)
- OrderFilled/ (monthly, already YYYY_MM.parquet) and CTF/ unaffected
Add daily_aligned_multi/ (neg_risk=true subsample, full history 2022-11-21 to 2026-04-28): 857 daily parquet files, 3.63 GB. Same cleaning pipeline as daily_aligned/ (relayer filter on both known relayer addresses + metadata join + p_event/D normalization via build_daily_aligned.py architecture), metadata filter switched from neg_risk=='false' to =='true', plus neg_risk_market_id column added (25 columns total). 100% full-history backfill, not incremental — every day verified against daily_aligned/'s date coverage before this upload.
