What problem does it solve? Agents and analysts routinely turn source material into claims without verifying whether the evidence actually supports each element of the claim, leading to overconfident findings, misgrounded citations, and unresolved contradictions entering reports and knowledge vaults. ## Core Features & Use Cases - Grounding Ladder Classification: Ranks every candidate finding from unsourced signal to independently verified finding, so only levels 4-5 can carry high confidence. - Element-Wise Grounding Checks: Breaks claims into actor, action, date, place, amount, and relationship, then maps each element to exact quotes, records, or metadata with source-role and support-type labels. - Confidence Caps and Contradiction Routing: Enforces maximum confidence rules (e.g., snippet-only sources capped at low) and classifies contradictions into value, time, existence, connection, and location categories without auto-resolving them. - Use Case: During an OSINT investigation, a finding states a company filed for bankruptcy on a specific date. The skill requires the exact filing record as a primary source, caps confidence at medium until independent corroboration exists, and emits a structured grounding object in findings.json. ## Quick Start Evaluate whether the cited sources in this draft finding actually ground each material claim element, assign a confidence cap, and flag any contradictions.
