The WASM boundary as a security primitive is underappreciated most people reach for containers without asking whether the threat model actually requires a persistent process. On the 10 strategies question: Supervisor and Mixture of Experts will cover 80% of real workloads in my experience, with Debate and Reflection worth keeping as the next tier since Debate surfaces assumptions single-agent pipelines miss and Reflection catches compounding errors cheaply. The part I would stress test is the export-to-Python parity keeping browser prototype behavior identical to the deployed artifact is a hard problem, and curious how you are handling edge cases where Pyodide’s stdlib coverage diverges from CPython.