Two mostly-independent changes landing together because they share the initramfs/firmware surfaces. Test path: oven build, ac-os pull, reflash. Phase 2 — kernel no longer embeds initramfs. It is now loaded externally by the EFI stub via `initrd=\initramfs.cpio.gz` in CONFIG_CMDLINE, from the root of whichever volume the kernel booted from. Effect: code-only changes repack one gzip instead of triggering a full kernel relink, and the two-pass monolithic+slim dance in docker-build.sh collapses into a single build whose output is the slim kernel (vmlinuz-slim kept as an alias for OTA compatibility). - kernel/config-minimal: CONFIG_INITRAMFS_SOURCE="" + cmdline initrd= - docker-build.sh: pack initramfs.cpio.gz up front, drop Step 4b slim pass entirely, ship the single bzImage as both vmlinuz and vmlinuz-slim - scripts/media-layout.sh: stage initramfs at stage root + copy to FAT / GPT image builders so USB / ISO / disk-image paths all include the external initramfs - scripts/flash-helper-runner.sh: copy initramfs to partition root for chainloader / kernel-only / kernel-direct / Mac HFS+ modes (systemd- boot already handled its own initrd= in loader/entries/) - pieces/os.mjs: OTA now fetches kernel then initramfs.cpio.gz, adds a downloading-initramfs state, passes both to flashUpdate - src/js-bindings.{h,c}: flashUpdate gains a third optional argument; flash thread also copies initramfs.cpio.gz next to BOOTX64.EFI with .prev rollback on failure G7 Chromebook (zesty-conch-resin log): - kernel/config-minimal: CONFIG_RTW89_8852BT + CONFIG_RTW89_8852BTE. Drawcia's WiFi chip is 10ec:b520, which matches rtw8852bte.c (HP- branded 8852BT combo variant with BT on PCIe function) — not the plain 8852BE I had enabled. drv=NONE on the PCI line in the USB log was the driver simply not knowing the ID. - initramfs/init: bump /dev/kmsg capture from 2s to 6s and snapshot post-probe state (net ifaces, /proc/asound/cards, PCI bindings). SOF's Jasper Lake machine driver probes late enough that the previous 2s window cut off right before codec/tplg load — missing the error that's blocking sound card creation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>