- Context for scoring LLM now includes evidence gathered during ICP validation
- lite_validation_node now strictly enforces tech_stack, unknown information is now rejected (previously only contradicting evidence caused rejection)
- Added more robust queries for LinkedIn profile search to cover more edge cases (profiles with the same name, similar roles at company)
Known issues: successfully parses but fails to query effectively for negative intent (eg. "e-commerce company with no app", "retail company with outdated tech stack"). More difficult to find lack of something from search engine.