Shared AI, local computation, and measured environments.
We start with open datasets, existing instruments, and questions small enough to test. Methods, code, raw observations, and limitations belong beside the result.
The work below describes proposed studies. Publications and datasets will be linked here as they become available.
Our first paper direction examines communal memory, attribution, permissions, and work that survives interrupted sessions. We plan reproducible evaluations of access isolation, source retention, and recovery.
Planned output: architecture paper and harness evaluation.
Read the directionWe plan studies of local inference, model routing, memory, and distributed execution under limited compute and connectivity. Open models and datasets let us compare latency, memory use, energy, and transparent baselines.
Planned output: rerunnable computational studies.
Read the directionStart with open data and a small maintainable set of measurements. Temperature, humidity, habitat recordings, and a site-specific water or soil channel form the initial station direction. An eDNA study is a separate sampling and molecular-assay collaboration.
Planned output: field protocol, acquisition dataset, and analysis.
Read the directionOur proposed mycelium study compares electrical recordings alongside measured humidity and temperature, with instrument controls and electrode-drift checks. Biomaterials are a separate track for moisture response and mechanical properties.
Planned output: controlled protocols and baseline datasets.
Read the directionA planned capability pack for literature, datasets, notebooks, analysis runs, and instrument records. It connects existing research tools through Aksara; approved findings can be shared independently or through an Strata community.
Tools & methods