mapping pipeline: review + land the design doc

[MJohnson459]

The design doc itself landed in #111 while this task sat in the queue, so the
work here was the review half: reconciling the doc with a month of shipped
work, indexing it from CLAUDE.md, and wiring the work breakdown to its Voro
tasks in both directions.

The doc against what landed

Four of the ten items have moved since the doc was written, and it described
three of them as pending and attributed one to the wrong task. Every item now
names its Voro task and takes that task's state as its own.

• Item 8 was credited to task 339. That is the preview pane; the candidate
  zone editor is task 346 (PR mapping pipeline: candidate zone editing from the dashboard #101). Stage 3's "edit before promoting" bullet
  carried the same mistake.
• Items 4, 7 and 8 read as unbuilt. 7 and 8 are in main (PR dashboard: promote picker cannot preview the candidate it promotes #100, mapping pipeline: candidate zone editing from the dashboard #101);
  4 is in review (PR mapping pipeline: map-build orchestrator #113), which also moved where it runs — a workstation
  beside bag_replay, not the fleet box, until item 5 gives a builder standing
  to upload — and its invocation is map-build -- --bag <bag>.
• Stage 3's "see the candidate" described an overlay on the promote picker.
  What shipped is a review mode that stands the operations panes down,
  because a canonical canvas with robots on it beside a candidate one without
  is the confusion the pane exists to remove.
• Items 3 and 10 gained their task numbers (342, 615); 2, 5, 9 gained theirs
  (341, 344, 347).

The substantive finding: stage 2, step 5

Carry-forward was written on zone/v0's vocabulary/binding split — names are the
stable half, coordinates are rebuilt per revision — and so made the step a
re-binding of names onto new geometry. Task 625 settled that the other way on
2026-09-02: a zone is a coordinate in the floor's frame, a fact about the
building. The kitchen does not move; a map is an estimate registered into that
frame. So a rebuild moves no zone at all, and the same-frame half of the step is
void rather than easier, while the cross-frame half becomes a question about the
map's registration and not the names'.

The step now states the reversal rather than quietly rewriting around it, since
625 has not landed. What survives is written out: match new rooms to existing
zones, propose the leftovers, emit the report the review pane reads. The "Stays"
row no longer claims the split stays, and the scope note points at 625.

Task 345 carries the rescoping: parked behind 625 (it would otherwise be written
against binding.yaml and rewritten immediately), with a note recording that its
acceptance also still named aliases and kinds, which task 491 retired.

Cross-referencing

Registered the doc as Voro doc 1 and linked the twelve tasks it spawned, so
voro doc show 1 is the progress view for the whole design and each task shows
the doc it derives from. Six task bodies still pointed at the unpushed branch
mapping-pipeline-design/af9cc81; those now name main and PR #111 (343 got an
appended correction rather than a body replacement, since voro show interleaves
its completion summary with the body and a clean extraction was not safe).

CLAUDE.md

A short section beside the fleet design references: the four stages, the two
rules that do most of the work (the live config is never deliberately hobbled;
a build gates on what it can measure and reports the rest), and a pointer to the
work breakdown as the state of the work.

Verification

• mkdocs build --strict passes — run twice, after each edit round. The doc was
  already in the nav from mapping design: the alignment step's lattice does not exist and its residual is unmeasurable #111, and the one path named that does not exist yet
  (docs/tuning/2026-09-01-map-build.md, banked with PR mapping pipeline: map-build orchestrator #113) is a code span,
  which docs/hooks/repo_links.py does not resolve.
• pre-commit run --files CLAUDE.md docs/design/mapping-pipeline.md clean.
• Prose lines wrapped at 80; the only overruns in the file are the pre-existing
  ASCII diagram and the stays/changes table.

No code changed, so there is nothing else to run. Branch
mapping-pipeline-land-340, commit b13dbec, not pushed.