How sources are scored
Source credibility is computed using a four-dimension geometric mean grounded in the epistemic philosophy of Miranda Fricker (2007) and Kristina Rolin (2011). Each dimension is scored independently; the geometric mean penalizes sources that score low on any single dimension, directly implementing Fricker’s sincerity criterion.
Scores across the source landscape
A range of representative sources, scored by the same methodology applied to every connector in the NormSense pipeline.
The range matters. NormSense does not weight all sources equally. An industry consortium publication is not treated as equivalent evidence to an enforcement action. Every observation in the system carries forward the credibility weight of its origin, all the way to the final fused score.
Defensible at every layer
Every observation in the platform traces back to its originating document, classified source type, and credibility weight. Every cluster requires observations from at least two independent source types before it is eligible for promotion to a tracked norm. Every Norm Confidence Score (NCS) is a Bayesian posterior whose inputs are listed and whose computation is reproducible. Every Fused Norm Evidence Score (FNES) is a weighted combination of streams whose weights are published.
This is what makes the platform’s outputs trustworthy to regulatory affairs leaders, in-house counsel, ethics committees, and policy researchers: the chain of evidence behind every signal can be inspected, audited, and challenged.
See it in Evidence Explorer
Inside NormSense, the Evidence Explorer surfaces the documents and sources behind every tracked norm. Open any norm to follow the chain from final score back to source credibility, and forward into adjacent observations and supporting documents. The provenance the methodology promises is the provenance the platform delivers.