Committed artifact · 2026-06-03

Performance evidence, with the boundaries attached

The current results are useful component measurements. They are not a claim that a Dits repository, media workflow, or future network runs at the same rate.

BLAKE3 hashing

1,809.96 MB/s

1 MiB input, 200 iterations

FastCDC chunking

991.76 MB/s

32 MiB input, 5 iterations

SHA-256 hashing

348.37 MB/s

1 MiB input, 100 iterations

Measurement environment

Apple M2 Pro, arm64 macOS, rustc 1.91.0-nightly, commit 9b79be227be8dd2cf1e9ead2a42e812ebf70565b. BLAKE3 and SHA-256 used 1 MiB inputs; FastCDC used 32 MiB inputs.

Not yet measured

The numbers buyers actually need

Dits still needs repeatable evidence for repository operations, storage growth, real-media fidelity, memory use, VFS latency, and competitive workflows. Network measurements are impossible today because network transfer is not implemented.

Required end-to-end suite

A result is publishable only when its inputs, method, machine, code version, raw artifact, and failures are visible.

A public corpus with generated fixtures and documented real-file characteristics
Controlled append, insertion, metadata, trim, reorder, and opaque re-encode edits
Add, commit, checkout, branch switch, integrity, and recovery operations
Wall time, CPU, memory, bytes read and written, and object-store growth
Cold and warm cache conditions plus byte-fidelity checks
Equivalent workloads for Git LFS, Xet, and relevant studio systems

Reproduce before you repeat

Run npm run bench from the repository root, then inspect the machine-readable artifact.

View raw results See the evidence roadmap