Only needed for free-text model IDs the backend cannot route by prefix.
OpenRouter LLM refinement is enabled. This improves uncertain boundary handling, but increases latency and token cost.
๐ก Tradeoff note:Force LLM refine + Use vision improve accuracy on edge-case filings (pre-HTML, plain-text, complex IBR), but add latency (~30โ120 s) and token cost (~$0.01โ0.05).
If you prioritise speed and cost over accuracy, uncheck both options โ the rule-only pipeline handles modern HTML filings at $0 cost in <5 s.