Release v1.18.62

v1.18.62

Stability

• Fixed the validator OOM / restart loop (health watchdog self-kill).
• Fixed an S3 storage write race that dropped score rows.

Throughput (~4×)

• URL dedup moved to a worker-process pool (GIL fix): batch wall ~21 min → ~6 min.
• Eval-scoped file cache reused across validation phases (~35–40% faster per eval).
• Eval batch raised 5 → 10 miners.

Data quality

• Miner file cap 512MB → 1GB.
• Reddit dedup now keys on immutable content IDs (post/comment ID), not the slug — closes URL-inflation. Handles /r/ and /user/ permalinks.
• URL↔content-id consistency check added to Reddit validation.
• Max duplicate rate 5% → 1%.
• Accept valid Reddit comments (OD metadata) and media-only tweets (S3) that were being wrongly rejected.

Community contributions — thanks to @afsarctg for #865 (Reddit comments in on-demand metadata) and #866 (media-only tweets in S3 validation).

Full details: #859