STRANGE and thoughtful AI research
current work
-

brightwoven
Active local model-training research: scaffold uptake, reasoning formation, correction, and observable training process.
-

evals
In development: continuity, state custody, boundary handling, false-route recovery, and model behaviour under pressure.
-

model behaviour
Ongoing fieldnotes on what models preserve, distort, suppress, overfit, route around, or reveal.
-

understanding
Strange outputs are not garbage. This lane treats them as evidence: traces of routing, pressure, uncertainty, correction, refusal, and repair.