SDC Historical PDF Reindex

What This Skill Debugs

This skill covers the recurring case where the workspace-level flag AI_INFRA__SDC_INDEXING_PARSE_FIRST_PAGE_MULTIMODAL was enabled after many contracts already existed, so older PDF contracts do not benefit from first-page multimodal stamp-certificate extraction until they are reindexed.

Observed recurring pattern:

• Support enables the flag and assumes the feature is now active for the whole workspace.
• Newer contracts behave as expected, but historical PDFs still show no metadata / no Stamp Certificate Number.
• Support or on-call has to manually:
  1. identify the historical ContractFullTextMetaData rows,
  2. soft-delete the stale rows,
  3. run ElasticBackfillReSyncRun in batches,
  4. verify extraction / repository visibility.

Confirmed repeat incidents:

• Zepto, October 2025
• Awfis, November 2025
• OYO, March 2026