What problem does it solve? Product teams assert claims about user needs without checking whether real people actually say those things. This Skill tests each claim from a research agenda against a corpus of real discourse, producing an evidence-backed Discovery document where every claim earns a verdict: supported, bounded, contradicted, blocked, or saturated-unevidenced. ## Core Features & Use Cases - Evidence-based claim scoring: Probes a discourse corpus through the eavesdrop CLI, judges each claim against a same-lens null control, and scores insights on five axes (frequency, intensity, friction, urgency, fit) with structured evidence citations. - Inverse pass falsification: Attacks its own supported insights with counter-probes at the authoritative pass, so an all-supported result is treated as unfalsified rather than confirmed. - Park-and-resume workflow: Parks as a draft when the corpus is missing or a question is open, arms a deferred wake via legion, and resumes scoring once the crawl accumulates. - Prediction instrumentation: Emits confidence-scored predictions per verdict and witnesses the intent review's claim predictions, building a calibration track record. - Use Case: After sd-intent-review produces a research agenda for a new auth product, invoke this Skill to test each claim against Reddit and forum discourse, surface emergent pain points the intent missed, and land a Discovery document with contradicted claims routed to the operator for rulings. ## Quick Start Invoke the sd-discover skill after sd-intent-review to test the agenda's claims against the eavesdrop corpus and land a Discovery document.
