BlackMark
5d520a1ff9
The flag was plumbed through pbrig.Deployment to the host-tool subprocess calls and nowhere else, so identity() and scan() opened a raw port and drove a shared line as though it were two wires. On real one-wire hardware the adapter's echo answers the knock before the device does, so the suite would have died at its very first check — "the loader never answered; nothing below can be trusted" — for the one deployment the flag exists to test, and every result after it is gated on that check passing. Both now open through pbrig.Rig.open_port(), which applies the deployment's link mode. The gap underneath was that only the subprocess path could reach those facts at all; anything driving the protocol in-process had to restate them, and did not. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>