Field notes and research from Raintree Technology on reliable AI systems, local-first web evidence, policy drift testing, agent costs, and reproducible workflows.
Field notes and papers, newest first.
Collect web evidence locally, enforce a zero-dollar budget, and preserve the results as inspectable Markdown, OKF, and SQLite artifacts.
Evidence: Boundary: --budget 0Boundary: --budget 0 | Benchmark: 120 provider runs
PolicyStrata detects cross-layer policy drift in LLM data agents with responsibility-scoped tests, clean controls, and minimized failure witnesses.
Evidence: Detected: 1,720/1,720 non-clean casesDetected: 1,720/1,720 non-clean cases | Clean controls: 80 with 0 false positives