Now
A mountable interface, if one is wanted. New scope rather than a debt — M2 asked for one facade and the S3 one shipped.
Backlog
Prioritized by what unlocks the next milestone.
All eight milestones are complete, so nothing here is owed — the backlog is what would make NexusFS reach further, not what would make it correct. It is grouped by execution order rather than by subsystem, and the cheapest item at the top is also the one that turns a tested rule into a working one.
Next
An incremental Merkle tree, and cached directory maps so a path lookup stops re-materializing each directory it walks through.
Later
A mountable interface, an incremental Merkle tree, and push notification so peers do not wait out the poll interval.
Research
Keep ZK, privacy, and advanced systems work feature-gated until the baseline is solid.
Highest-priority backlog
- See through a VPN, and spot a Wi-Fi hotspot on macOS — the link costs detection still cannot read
- Prioritise which deferred content to fetch first under a capped budget
Recently cleared
- Per-recipient key envelopes, so replicas share no repository key
- Key rotation, so removing a peer withdraws access from then on
- Push notification — 0.32s to a peer, against a 60s poll interval
- Metered-link detection, so the scheduler's link rule can fire
- A storage-headroom floor replication will not cross
- Collection of orphaned namespace records, not only blobs
- The energy budget in
nexusfs status, not only over HTTP - On-demand fetch — a deferred file is fetched when someone reads it
- Incremental snapshots — the inode map is maintained, not re-walked
- One fsync per operation, not two for the same database
- Absence proofs, and the forged-length attacks they must refuse
- Peer enrolment, format migration, and garbage collection
Planning
Roadmap and backlog now work together
The roadmap defines milestone outcomes. The backlog defines the concrete queue that moves the project toward those outcomes.