Set USAGE_EVIDENCE_BASE and USAGE_EVIDENCE_HEAD to the two installed checkouts. Build the server bundle and web client in each, copying web dist to server dist/client. Run scripts from the head checkout; keep these scripts together in a scratch directory. prepare.py creates the synthetic fixture. batch.py runs controlled cold and warm RPC comparisons. capture.py records Usage in local Playwright. edges.py and append.py run real-server failure and append exercises. reopen.py targets the recorded batch state directories and therefore needs its run directory adjusted for a new run. EVIDENCE_EXECUTABLE optionally selects the Electron executable; launch.py then sets ELECTRON_RUN_AS_NODE=1. All output is synthetic. Do not upload config-private, cookie-private, pairing files, raw browser summaries, or raw server logs. Public report files omit environment fingerprints and replace source paths with synthetic labels. The capture script adds a timing annotation; no application source was modified.
