WORK-001: bare `work` shows the worktree forest

Now: DONE

Input

Context

Bare work (BE-043, unlanded board in work/BE-043) lists a flat button row per work/ worktree; the bare form must instead render the FOREST: the main tree with its mounts, the branch trees, and a foreign tail — every worktree hung under what it TRACKS.

Fleet survey 2026-07-18: 65 wts under work/; 34 anchor file:~/.be/?/beagle-ext, 7 file:…/jab/.be/?/libabc, 9 foreign (libdog etc); most latest get rows are detached ?#hash.
Track classification exists: shared/wtlog.js attachedBranch() → { uriTrack, track, br, detached, base }; sub pins are self-pointing get //wt/path/sub#pin rows (DIS-072, Submodules).
DAG ground: shared/dag.js ancestors/aheadBehind walk per pair today; shards and the nested branch-name tree per Store.
Models: views/status mount recursion (gitmodules order), views/du store scan, BE-043 views/work board (workDir, buttons).

Goals

Bare work renders three blocks — ASCII tree in plain mode, HUNK spans + hidden U nav in the pager:

the TREE: the top wt as root, mounts recursed in .gitmodules order; each line <date5> #<hashlet8> <subject> off that line's cur tip; a work wt tracking a wt pin hangs under the tracked wt.
BRANCHES: store → shard → branch (nested by name segments) → the wts tracking that branch, named by authority path (e.g. //WORK-123/test).
FOREIGN: refs/anchors outside the project's stores, with the wts tracking them.

Every wt row carries ahead/behind counts vs its TRACKED ref (ruling below), rendered compactly after the subject.
Siblings name-sorted everywhere; a work wt is ONE row — no descent into its own mounts.
DAG cost batched per shard: keeper opened ONCE per (store,project), ancestor closures memoized per distinct tip — hundreds of wts share a handful of tips.

Constraints

Read-only view; discovery = readdir of work/ (the BE-043 .be-file gate) + the context tree's mounts; NOTHING hardcoded.
ALL URI/track/anchor parsing via the wtlog reader / uri / core resolve — no regexes, no string surgery (work §NOT TO DO).
Date column: the existing ron.date / render.dateCol 7-col utility (the 5-char 18Jul form) — no new formatter.
Keep BE-043's machinery (workDir, buttons, tests) — the forest is the BARE form; the flat board's chrome rides the wt rows.

WIP

Design decisions

Linking model superseded 2026-07-19: the forest links by TRACKING edge (tracked == upstream), work wts nest recursively — WORK-007; the mounts-skeleton one-level attach below is stale.
RULING (chat 2026-07-18): ahead/behind is computed against the wt's TRACKED ref, never the shard trunk or a shard-identity guess.
RULING (chat 2026-07-18): link by TRACK — wt-pin trackers under the tracked wt (block 1), branch trackers under the branch (block 2), foreign refs + their trackers block 3.
RULING (chat 2026-07-18): siblings name-sorted; the forest is part of work (bare form), not a separate verb.
RULING (chat 2026-07-18): the blocks are TREE HUNKS — main-tree-tracking, branch-tracking, remote-tracking; an empty hunk is ABSENT, not rendered as an empty header.
RULING (chat 2026-07-18): rails are Unicode box drawing (├── └── │), not +-- ASCII.
Approved sketch off the live fleet (grep get work/*/.be, 2026-07-18; 60 wts, all detached pins → branch hunk absent; subjects/counts elided):

16Jul #........ journal — the top wt
├── be                18Jul #226c6228 …        beagle-ext ⇐ ~/.be
│   ├── //BE-043          14Jul #33295788 …
│   ├── //PATCH           18Jul #226c6228 …
│   │   … (34 tracker wts, name-sorted) …
│   └── test          09Jul #........ …       beagle-ext-tests
│       └── //beagle-ext-tests  09Jul #73fc0c91 …
├── jab               16Jul #016be1b3 …
│   ├── //JS-114          11Jul #e300c639 …
│   │   … (7 wts) …
│   └── dog/abc       17Jul #........ …       libabc ⇐ jab/.be
│       └── //ABC-017     17Jul #8d40ba3b …
│           … (7 wts) …
└── //BLAME-001       10Jul #5041ed18 …       tracks the top wt

file:~/src/libdog/.be ?/libdog                remote
├── //DOG-008         10Jul #79612854 …
│   … (9 wts) …
file:~/src/jab.be/.be ?/jab                   remote
└── //DOG-021         14Jul #b58be37a …

Fleet oddities the classifier must survive: //SUBS-054 anchors file:/home/gritzko/.be/beagle-ext (no ?/); the seven JS-* wts anchor the jab WORKTREE (file:…/journal/jab?/jab), not a store.
RULING (review 2026-07-18): main-tree repo rows (root + mounts) render BOLD in the pager; tracker wt rows unstyled; --plain carries no styling at all.
RULING (review 2026-07-18): wt row layout = //KEY [get] [post] <ahbeh> <time5> #<hashlet8> <subject> — buttons named for their VERBS (update→get, merge→post) and placed right after the key so they are always reachable; the ahbeh field is padded (~8 chars), behind in salmon, ahead in salad green.
Review 2026-07-18 confirms: project root as the first hunk's root is right; an empty branch hunk absent is right (zero wts track a branch today).
RULING (review 2026-07-18): [diff] rides BETWEEN [get] and [post]; button colors blue/yellow/green respectively; the subject is trimmed to ~30 chars; the key/URI click navigates to the wt's status.
RULING (review 2026-07-18): linking — a wt tracking a mount/tree hangs under that tree; a wt tracking a parent's PIN hangs under that parent; a wt tracking a BRANCH goes to the separate store → shard → branch → //worktree tree (hunk 2).
RULING (review 2026-07-18): tracker rows hang on DOTTED rails (┄), actual subdirs/mounts on solid ─ — so trackers never read as real subdirs. CONFIRMED (gritzko 2026-07-18): mounts first (solid, .gitmodules order), then ALL of the node's trackers as ONE name-sorted dotted run (supersedes the grouped trunk-pin/wt-pin runs).
RULING (review 2026-07-18): the hunk-2 tree is store → shard → [branch →] //worktree — the branch level is OPTIONAL: a detached wt whose shard maps to no mount (//BLAME-001 on shard project) hangs DIRECTLY under the shard node, branchless. An anchor may omit the shard name (bare ? trunk / the ?/-less SUBS-054 form) — resolve it through the standard resolver (never guess); the wt always lands on a concrete shard.
RULING (review 2026-07-18): [rm -rf] is replaced by [done] and [dont], placed AFTER the message: each MOVES the wt (recoverable discard — no safety gap), and a TICKET-named wt also flips its ticket header status to [DONE] / [DONT] (page via the shared ticket lookup); a non-ticket-named wt just moves. Supersedes the rm //KEY emit and reshapes BE-044's verb into move+mark.
RULING (review 2026-07-18 r2): the discard root is work/done/ — done/dont wts mv there (same device, no EXDEV; supersedes /tmp/<basename>), and the view IGNORES work/done/ entirely.
RULING (review 2026-07-18 r2): main-tree repo rows embolden the NAME only, the rest of the row normal weight.
RULING (review 2026-07-18 r2): the name/key column pads RIGHT to a fixed ~16 chars with a dotted filler line, so every row's buttons start at one aligned column.
RULING (review 2026-07-18 r2): main-tree rows' time/hashlet/message columns align to the SAME offsets as wt rows' (repo rows pad across the buttons+ahbeh region).
RULING (review 2026-07-18 r3, amended): the ahbeh cell is _+99-99 — content right-aligned to 7, ONE space each side (the right-side gap is dateCol's own lead, no trailing pad); both counts inline; display clamps at 99.
RULING (review 2026-07-18 r2): main-tree repo rows MAY show ah/beh vs their own tracked ref — for a mount that is likely the PARENT's pin of the sub (the de-jure sub pin), rendered in the shared ahbeh column.
(pending) a detached wt (bare get ?#hash, tracking neither wt nor branch) hangs under its anchor SHARD node in block 2?
(pending) does the branch block list ALL shard branches or only the tracked ones?
(pending) do block-1/2 wt rows keep the BE-043 update/merge/[rm -rf] buttons in pager mode?

TODOs

red-first fixture forest (test/work/view/): main wt + nested mounts + four wt flavors (pin-track, branch-track, detached, foreign) → pinned plain output
classify: work/ scan + attachedBranch() → { block, parentKey } per wt; anchors resolved through the wtlog reader only
shard registry: (store,project) → keeper opened once; per-tip ancestor-closure memo; aheadBehind COUNTS vs the tracked ref (shared/dag.js)
model: the three blocks as pure data, name-sorted; branch nesting off the shard refs rows (branchlib)
render: plain tree with Unicode box-drawing rails (ruling above); pager HUNK spans — U nav per row, buttons per the pending ruling
wire as bare work; the flat board absorbed; BE-043's tests updated to the new bare form

Blockers and bummers

Landing 2026-07-18 stalled on ONE pager.js weave conflict → POST-032 (conflict must mark+record, not hard-err; carries the SALVAGED hybrid resolution — merge it into this wt's pager.js before re-posting; local commit c32ddd17 holds the full change).
BE-043's work.js is UNLANDED (lives only in work/BE-043) — land or absorb it first, else two work bare forms collide.

WORK-001: bare work shows the worktree forest