Diagnosed from vital-shark-husk USB log: G7 (Jasper Lake, Drawcia) exposes Realtek RTL8852BE WiFi on PCI 01:00.0 + SOF DSP at 00:1f.3 with "no soundcards", plus install-to-disk targeted /dev/mmcblk1p12 after sfdisk had already rewritten the partition table to a single p1. - Dockerfile.builder: add `realtek-firmware` — Fedora 43 split rtw89/*.bin out of linux-firmware into a dedicated package, so rtw89 driver loaded but its firmware_request() silently failed (drv=NONE in the log). - docker-build.sh: recurse into intel/sof/{community,intel-signed}/ when staging DSP firmware. Fedora's alsa-sof-firmware ships sof-jsl.ri under subdirectories, not flat, so a top-level glob missed the Jasper Lake blob and SOF platform probe found no card. - kernel/config-minimal: reboot=efi,cold — some Chromebooks hang when the kernel tries its default reboot path (ACPI then KBD then triple- fault). `reboot=efi` falls through to UEFI ResetSystem() which works from any EFI-booted kernel; `cold` is the final fallback. - ac-native.c install: after sfdisk shrinks the Chromebook's 12-entry GPT to a single 1024MB ESP, the new partition is always p1 of the parent disk — canonicalize devpath before waiting on the node, mkfs, and mount. Previously we kept `/dev/mmcblk1p12` and waited 10s for a device that no longer existed, hit "format failed!" banner, then recovered on pass 1 which made the telemetry look alarming even though the install eventually succeeded. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>