Bulk reprocess a profile¶
Re-extract all sources compiled by a specific profile version. Useful when you've improved extraction prompts and want to recompile the entire corpus.
Basic usage¶
This:
- Finds all sources associated with the profile
- Re-runs the full ingestion pipeline for each source
- Produces a summary with triples added, removed, and unchanged
Dry-run mode¶
Preview what would be recompiled without executing:
Limit the scope¶
Recompile only the first N sources (useful for testing):
When to recompile¶
- After updating the profile's
prompt_text - After upgrading the LLM model (e.g., switching from
gpt-4togpt-4o) - After fixing an extractor bug
- After expanding
competency_questionsand wanting to verify the full corpus
Cost considerations¶
Bulk recompile triggers LLM calls for every fragment (hash check is bypassed). Estimate the cost first:
Look at the "Cost" column to estimate per-fragment cost, then multiply by fragment count.
Warning
A full recompile of a large corpus can be expensive. Consider using --limit to test on a subset first.