Tracks are the substrates we work on — water, food, frontends, epistemics, methodology. The thesis and the investigations cut across all five. Tracks are the “what we work on” axis; the thesis is the “why”; the investigations are the “applied here.”
Community-operated monitoring + harmonized open datasets + physics-informed AI for environmental systems.
Starting with the Memphis Sand Aquifer: an open IoT sensor network, a 30-source harmonized canonical dataset, and a physics-informed graph-attention architecture with aquifer-unit priors. The drilled substrate already exists — 4,257 wells in the metro, fewer than 53 streamed continuously in the open. What's missing is the open monitoring layer.
Small-footprint, automated indoor food production; reframing what "required" nutrition means under longitudinal household evidence.
Not agriculture, not industrial kitchens — a research substrate at the scale of a family. What can be grown, at what nutritional yield, under what automation, in a spare room or a balcony? And what does a measured self-sufficiency layer change about resilience, nutrition, and the political economy of food access? Outputs: open-hardware designs, measured yields, and evidence-informed reframing of inherited nutrition defaults.
Research into how adaptive, multi-modal, multi-depth, multilingual interfaces change who can access expert knowledge.
The frontend of learning has always been the constraint — not intelligence, not effort, not the availability of knowledge. Post-industrial schooling built one linear filter and called those who fit it "smart." This track tests whether building the plural frontend changes the distribution of who can learn, measured honestly. The Echo-family products (Tales, Learn, Birds) are the current test substrates.
How does a society retain the capacity for collective error-correction when the dominant knowledge frontend is commercially governed and silently versioned?
Two levels. Engineering: confidence tags, citation binding, refusal-to-fabricate, red-team transcripts — patterns that keep AI systems honest about uncertainty. Ecological: longitudinal observatories of production-model drift, provenance layers for AI-mediated claims, methods for preserving diversity-of-error. This is the track that makes LeResearch worth existing as a distinct entity.
What an insight from one substrate teaches us about another. The track that keeps LeResearch coherent.
The silo-collapse thesis applied to LeResearch itself. Methodology papers, pattern libraries, annual cross-substrate meetups, annotated comparisons of the same methods across different fields. If the thesis is that domains connect, this track is the receipts.
The companion browsing axis is /topics — the conceptual cross-cuts (capacity, normalization, labor, surveillance, monoculture, …) that span the substrates.