The Journey of One Operation

Architecture is the map. This is one trip across it.

The scene. bl0k is a ticket you just finished; fr33 is waiting for it. You close bl0k — and fr33 becomes the next thing to do, without anyone writing that down.

You close one ticket

nxf close bl0k --reason "shipped in #412"

The CLI holds no model of its own. It calls Engine::close(now, actor, id, reason) on crates/facade — the same seam an embedding app calls. There is no private path the CLI can take and an app cannot, which is why CLI coverage never counts as proof for the seam.

The log grows by three rows

write::close appends rather than writes: set_field(status), set_field(closing_comment), set_field(closed_at). Each append becomes one row in the ops table.

12  bl0k  status           in_progress        ← still there
13  bl0k  status           closed
14  bl0k  closing_comment  shipped in #412
15  bl0k  closed_at        2026-09-02T09:14Z

Nothing was overwritten — and that holds for every verb before this one. nxf create appended five rows per item, nxf dep add one edge row, nxf claim one more. The log is the history: every op carries its own author and wall_clock, so who changed what is derivable from it without a second audit table to keep in step.

The views are refolded

A reducer folds each new op into the items view under keep-if-beats LWW — an op wins only if its (lamport, site) beats what the view already holds, so two machines folding the same ops in different orders land on the same row.

The view is a cache of the log. Delete it and a refold rebuilds it, which is why sync ships the log and never the view.

The answer changes

$ nxf next
showing 1 of 1
bl0k  P2  in progress  [feature]  Ship the export path

$ nxf blocked
fr33  P2  open         [feature]  Publish the diagram
    ↳ blocked by: bl0k (in progress)

$ nxf next
showing 1 of 1
fr33  P2  open         [feature]  Publish the diagram

$ nxf blocked

The second nxf blocked prints nothing at all — the lane is empty, so there is no line to show. Between the two runs there was nothing but those three appended rows. blocked_gating (crates/core/src/derive.rs) holds an item back only while a dep is not yet closed, so fr33 stops matching: the predicate ran again, nothing wrote to it.

No row anywhere says which item comes next. nxf next and nxf blocked are questions, asked fresh each time — not columns someone keeps up to date. That is what makes the board consistent without anyone maintaining it, and it is what the schema pays for the privilege: whatever you want to ask has to be derivable from the log.