Where it actually is
v0.22 is the research engineering baseline. The web-based AI that produced v0.1 through v0.22 has handed off; local work starts from here.
How it grew
Container, profiles, Merkle basics
Content-addressed store, copy-on-write, packfiles
Chunking, indexes, packsets, background maintenance
Transparency log, witnesses, public checkpoint service
Supply chain, TUF, PKCS#11 semantics, evidence governance
The paper and the implementation have diverged
The whitepaper is still at v0.1 while the implementation reached v0.22. The specification documents now cover B+ trees, LSM compaction, TUF, PKCS#11, HTTP/3 workers and transparency logs — a storage and supply-chain governance system rather than a text container. The .u8x interchange format itself has not changed since v0.21.
This is worth deciding rather than drifting through: either bring an encoding paper up to v1.0, or acknowledge that the project became something else and give that a paper of its own.
What the hand-off asks for next
Not more features. The hand-off document is explicit that the next stage is convergence — making the project installable, reproducible on another machine, and honest about which external dependencies were actually exercised.
- Make it run to completion on a second machine. Currently it does not.
- Turn the pipelines that were never really executed into real results: protoc, aioquic HTTP/3, Rust wheels, genuine public Sigstore and TUF corpora.
- Choose one product direction and keep the rest as research modules.
- Measure random-access amplification, so the architecture can be judged on the axis it is designed to win.
History is the git history
All 21 archives are replayed as one commit per version with a tag each, so any file from any version is one command away. The v0.6 archive is missing from the source folder, but its work is not lost — its report, notes and eleven passing tests all survive inside later versions.
Six CJK filenames needed repairing on the way in. v0.21 stored them double-encoded while setting the flag that asserts they are correct UTF-8 — worth recording plainly in a project about lossless UTF-8 round-tripping.