Active correction audits an initial retrieval result and intervenes only when it is inadequate. A retrieval evaluator sends low-confidence results through a corrective web-search loop, targeting staleness and single-fact hallucination. section 11.4
Use active correction when retrieval reliability matters more than unconditionally expanding the search:
- Keep the fast Correct path for adequate results; trigger web search only on the Incorrect path. section 11.4
- Apply it to fact-checking and real-time validation, such as verifying whether someone meets UK PEP rules. section 11.4
- Combine it with proactive planning when queries require decomposition, multi-hop logic, or heterogeneous retrieval. section 11.4
Table 11.1 Strategic selection matrix
|
Feature
|
CRAG (Corrective)
|
LevelRAG (Plan-and-Refine)
|
|
Primary Mechanism
|
Evaluation & Fallback
|
Decomposition & Hierarchical Search
|
|
Best Use Case
|
Fact-Checking, Real-time Validation
|
Complex Research, Multi-document Synthesis
|
|
Key Innovation
|
Decompose-then-Recompose
|
High-Level Logic Planning
|
|
Compliance Role
|
Screening ("Is X a PEP?")
|
Advisory ("Compare Policy A vs B")
|
|
Latency Profile
|
Low (Internal) / High (Web Fallback)
|
High (Iterative Planning)
|
|
Architecture
|
Component Add-on
|
Full System Architecture
|
Compare active correction with hierarchical planning across mechanism, use case, compliance role, latency, and architecture. section 11.4
