Lab 14 — Cross-harness comparison

Use this lab’s provider.env.example and the preflight. Run one interactive, fresh Copilot session in an empty disposable workspace. Ask for a Store Health Report with only substantiated service/catalog claims and no claimed fixes without a write log. Record version, provider (GitHub-hosted versus Foundry BYOK), model/deployment, tool calls and sandbox/approval settings. Note unavailable provider usage as unavailable rather than zero.

An apples-to-apples scorecard requires the same pinned source, seeded simulator, task, model deployment, tool access, run count, grading rubric and external write audit for both harnesses. This path does not yet provide that integration; do not compare a GitHub-hosted run with Foundry as if the model were identical, and do not claim an empty workspace disables the native loop.