Human–AI interaction raises a question: how can people inspect the evidence behind fluent AI responses? This research perspective connects explainability, conversational memory, uncertainty and human judgment. Founder-led Epistemic Trust Lab is a separate local instrument for deterministic synthetic evaluation, not a human-trust study.
Developer-reported local build counts · not human trials · not real-model benchmarks
Nurowerx founder Spencer H. Walker has implemented a local research instrument for studying conversational memory, evidence support, apparent AI understanding, and measured reliability. It includes a memory/evidence inspector, quantitative metrics, uncertainty intervals, exports, and a clean-install reproduction.
240 synthetic scenarios and 5,760 executed deterministic policy evaluations. 34 tests reported: 29 backend, 2 frontend, and 3 browser, including automated accessibility checks.
These figures are developer-reported; the code archive has not yet been independently audited online. Synthetic policy results are not evidence of actual language-model behavior or human trust effects. Real-model validation and authorized human-participant research remain future work.
Read project research overview →Founder research note →The memory/evidence inspector connects the research question to source attribution, contextual inference and visible uncertainty. Metrics and uncertainty intervals make the reported synthetic-policy outcomes inspectable; they do not measure what actual people understand or trust.
Nurowerx explores interfaces that preserve human context, allow evidence review and support correction or override. These are research directions, not demonstrated improvements in human decisions or validated explanations from real language models.
An Epistemic Trust Lab read-only import into Research Stack was developer-reported as verified. The original ETL project remained unchanged; importing its data does not revalidate it or add human-participant or live-model evidence.
Local OLMoE inference with router-faithful expert residency. Compare the v0.2 memory, latency and fidelity findings with their experimental limits.
Public research software v0.1.0: TRACE · LOOP · LENS · FRAME connects provenance, evaluation, visualization and reviewed corrections. Includes NRS-Bench, three technical reports and 54 developer/CI-reported tests on controlled synthetic evaluations, not a real-model or human study.
Recorded-video detection, temporal tracking, replay and exports in a local prototype. The synthetic map is not verified geolocation.
Public notes disclose reported methods and limitations; they are not a substitute for independent validation.
All research themes →