Continuous, multi-source signal capture
Norms governing advanced systems do not emerge in any single venue. They form across regulatory rulemakings, court decisions, standards body deliberations, executive orders, industry self-regulation, academic critique, civil society advocacy, and shareholder action — each with different velocity, credibility, and lead time.
NormSense ingests across all of these layers continuously, on a six-hour cycle. Sources include federal and state legislation, SEC filings, Federal Register rulemakings, lobbying disclosures, academic papers, think-tank publications, civil society reports, and standards-body drafts.
Norm observations, not document summaries
Each ingested document is processed for norm observations: discrete claims about what should or should not happen with advanced systems, who promotes or requires the norm, and how it relates to human agency. Observations are tagged with the dimensions of agency they concern, the actor advancing them, and the lifecycle stage they reflect.
The extractor distinguishes between negative norms (widespread practices that erode agency), positive norms (governance standards that protect agency), and emergent patterns (positions not yet settled in either direction).
Observations resolve into norms
Individual observations are clustered using semantic and lexical similarity into coherent norm units. Each cluster represents a distinct normative claim, accumulating evidence over time and across sources. As more observations accumulate, the lifecycle stage of the norm — emergence, contestation, crystallization — becomes measurable rather than asserted.
Every claim traces to its evidence
Every norm in NormSense is auditable to its underlying observations, and every observation traces back to the original source document with byte-level provenance. There are no black-box assertions. The pipeline is designed for the kind of work where explanation matters — regulatory affairs, in-house counsel, ethics committees, academic research.