Verification
A copy that finished is not the same as a copy that is correct. Cocreate hashes during the read, checks the result off the disk rather than out of the cache, and only then puts the file in place.
- Scan214 files · 1.42 TB
- Copyread once, hashing as it goes
- Verifyre-read from the disk
- Publishatomic rename into place
- Receiptwritten
Cancel at any point and nothing half-written is left behind.
How a transfer runs
- 01
Scan the source
Cocreate walks the card in parallel and tunes how hard it reads to the storage underneath — four workers on a spinning disk, sixteen on NVMe, because too many readers make a card slower rather than faster. Unreadable and ambiguous files are reported before a byte is copied.
- 02
Copy and hash in one pass
The source is read once. Every checksum you asked for is computed from that same read, rather than copying first and hashing after.
- 03
Verify off the disk, not out of memory
Bytes are staged in a hidden folder beside the destination and flushed, then read back off the disk and checked. A cache reporting success it cannot back up gets caught here.
- 04
Publish atomically
Verified files are moved into place with a rename on the same volume. Cancel halfway through and there is no half-written file to find later — only staging to clear.
- 05
Write the receipt
Each roll gets a receipt recording what arrived, what it hashed to, and anything that did not match.
Choose how sure you need to be
Pick as many methods as the job warrants. They all come off the same single read of the source, so a second choice costs far less than a second pass would.
- Fastest
File size check
Compare file sizes only
- Fast
CRC32
Fast error-detection hash
- Fast
xxHash
Extremely fast non-cryptographic hash
- Medium
MD5
Standard checksum algorithm
- Medium
SHA-256
Industry standard cryptographic hash
- Slow
SHA-512
Stronger cryptographic hash
- Slow
Byte-by-byte
Binary comparison, the most thorough there is
Work it does not repeat
Checksums it already has
Where a checksum exists in the project, in a manifest at the card root, or in a per-file sidecar, Cocreate uses it instead of computing it again — and reports how many it reused against how many it had to calculate.
Picks up where it left off
Each file is an independently restartable unit, so an interrupted job resumes at the next file rather than starting the roll again. Files already verified are not recopied.
Knows a retry from a failure
A share that dropped, a file another app had locked and a scanner holding a handle are treated as worth retrying. A permission problem or a full disk is not, and is reported straight away.