The ECC-safe restore is now proven end-to-end on a bench S17: stock -> Alkane 0.1.3 (v2 OOB backup carved) -> network restore-from-onboard -> genuine stock booted clean (lighttpd 401 "antMiner Configuration"), no ECC-unrecoverable brick, no SD card. Two fixes found on hardware: - This mtd-utils `nanddump` defaults to --omitoob and to ECC-correcting. The restore read-backs (scratch self-check and per-slot verify) must read RAW, WITH OOB (`nanddump --noecc --oob`): the boot slots carry the foreign stock ECC, which the alkane soft-BCH engine would "correct" into garbage, and the default omit-oob made the read-back shorter than the written data+OOB. Proven on the board that `nandwrite --noecc --oob` + `nanddump --noecc --oob` round- trips the stock OOB byte-for-byte. - carve `dump_with_oob` now passes --oob explicitly instead of relying on the nanddump default (version-robust OOB capture). Also adds a method-free progress logger (`$ALKANE_OUTCOME_LOG`) so a sealed, DSH-driven from-stock install can report its own steps (root/carve/write/verify) without relying on the harness to summarise - the lines are written by this code and never carry the stock-access method. 93 flash-crate tests green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>