Deep Escalation

Deep tier escalation moves 'needs_review' records into a multi-round LLM investigation with web search integration. It specifically targets records stuck after deterministic and AI planning phases of a data cleaning pipeline.

Reusing prior web-search evidence stored on the record avoids redundant web searches during LLM record escalation. The deep tier investigation checks existing evidence first before initiating new external queries.

Budget-gated execution for deep tier LLM data cleaning restricts multi-round investigation to run only when necessary based on allocated budget and record status. This prevents excessive API usage on records not requiring deep analysis.

The deep tier LLM automatically logs audit decisions and flags directly onto the record after escalating records for deep analysis. This occurs once the multi-round investigation concludes and a final verdict is reached.

You should not use deep LLM escalation for data cleaning if records have not passed through deterministic and AI planning phases, or if budget constraints cannot support multi-round investigations with web search integration.