Forensis-Lab

One nibble: when 0xC4 stopped looking like broken Ford

4 August 2026 is the day a complete, clean backup download stopped being a fantasy. For months we treated lead byte 0x04 as the sacred marker of “this host is Ford / this slice is decryptable.” Force paths scanned for 0x04; anything else was discarded or forced through Ford open checks that could not succeed.

The fulcrum was absurdly small: many data ChunkInfo bodies do not start with 0x04 at all. They start with values such as 0xC4 (yes—a C where we insisted on a 0). Those slices are still valid ciphertext for the chunk pipeline; they are simply not Ford CT. Once the classifier stopped requiring lead-0x04 for data gates—and kept 0x04 as the typical Ford CT marker—the remaining thousands of files fell into place. Side-by-side with commercial tools, reconstructions covered nearly the whole inventory; the gap had been a nibble, not a missing Apple API.

Completeness still has honest ceilings. Some small files are removed from CloudKit by Apple or third-party developers: the backup inventory may still name them, but there is no longer an MMCS record / authorizeGet asset to fetch—you cannot invent edge bytes that the account no longer holds. Separately, a few payloads (notably some Mozilla translation / on-device AI model style files we hit in testing) are encrypted under schemes that are not the Apple Ford/MMCS path we reverse-engineered. Those remain opaque without a different oracle; calling that a “failed MMCS decrypt” would be a lie.

So August unlocked a protocol-complete Apple backup download for the material that still lives on MMCS—not a claim that every historical filename on earth is recoverable.

← All notes