Beagle SCM

WORK-013: work view tree-row clicks are relative U words — from a non-root context they NAVNONE or show the wrong tree

Reported by gritzko 2026-07-19, live: jab work run from //WORK — clicking the main tree's abc mount row (jab/dog/abc) drives //WORK: status jab/dog/abc → NAVNONE; clicking the root row drives a bare status → shows //WORK's status, not the main tree's. The rows bake context-RELATIVE U word spells at render time; under any pager context but the project root they resolve against the wrong tree. Must catch up to the BE-054 O ruling: verb clicks are O spells, U is reserved for real addresses, and tree rows carry FULL ///path/ addresses.

Input

Context

• Mint sites, views/work/work.js treeRows: root row nav: navlib.navLink("status", "") (:646) → bare status; mount rows nav: navlib.navLink("status", c.relpath) (:666) → status jab/dog/abc — relative, no authority when the render-time nav authority is empty (shared/nav.js:48-55 composes off authority(), not the row's own tree).
• wt rows are half-converted: buttons mint O (SPELL.mintOspell, work.js:491-567) but the row-key/± clicks still bake U words status //KEY / diff //KEY (work.js:530,544) — authority-full so they resolve, yet still U verb-words the BE-054 ruling reserves for real addresses.
• BE-053 (landed c2768ce7) keeps the pager context on a word-spell arg click — so the bar stays put, but the VERB still resolves the relative arg against the pager context: //WORK has no jab/dog/abc → NAVNONE in plain words; root's bare status = status of the current context.
• Rulings in force: BRO-025 three-part O (context · verb · args); use URIs to name trees (///sub/, //WT); resolution via core/resolve_hash.js only.
• Live refinement (gritzko 2026-07-19, rechecked from //WORK-013 — launch-independent): the abc click lands on //<LAUNCHWT>/jab/dog/abc/: status, bar RED, view = the launch wt's status — navLink BAKES the render-time authority() onto the mount's relpath (nav.js:48-55), minting a real-looking WRONG-tree address the pager then legitimately navigates to. Not a BE-053 regression — the weld is in the mint, not the pager.

Goals

• Tree-node rows mint context-independent O spells: root → status ///, a mount → status ///jab/dog/abc/ (or the O context slot carrying the tree, per the BRO-025 wire form) — a click works identically from ANY pager context.
• wt-row key and ± clicks convert U verb-words → O spells (status //KEY, diff //KEY as O), finishing the BE-054 catch-up in work.js; after the sweep the work view emits no verb-word U tokens.
• Red-first pty repro: jab work launched from a NON-root worktree cwd, click a mount row → the mount's status renders (no NAVNONE, no wrong-tree status), pager context unchanged.

Constraints

• Mint via SPELL.mintOspell only, addresses via the standard composers — no hand-built URI strings; do not touch views/bro/pager.js.
• U tokens that ARE real addresses stay U; the fade #rrggbb O and button Os stay as-is.
• test/work/view and test/work/getctx pin current row bytes — amending their pins to the O form is AUTHORIZED; no other test edits.
• Overlaps WORK-014 (mount ahbeh vs branch tip, same treeRows region) — coordinate, separate commits; build on //WORK/ tip, expect weaves.

WIP

Design decisions

• gritzko O-context ruling (2026-07-19): O = context·verb·args; context EMPTY=leave, ABSOLUTE(/…)=REPLACE, RELATIVE=combine; ARGS go to the verb VERBATIM (pager never welds/reduces them).
• Tree rows: ABSOLUTE context ///<relpath>, verb status, NO arg (nodeCtx) — root ///: status, mount ///vend/ext/: status.
• wt rows: context-less O, arg VERBATIM — status //KEY, diff //KEY (U verb-word → O tag; //KEY is a full address).
• prefixSpans nav span is TAG_O but PAGER-ONLY: the nav O is minted only on the btns path (wtSpans + emitRows repo row gated on btns, mirroring fadeHex/BE-054) — an O in a PLAIN hunk trips bro.js's why-plain cursor and drops the work banner; plain stays byte- identical to the golden. navlib(navLink) require dropped — no verb-word U left in work.js.
• GAP (bro-side, NOT fixed per order): parseOspell strips a bare /// context to "" ("here"), so the ROOT click drives status in the EMPTY context (statuses the pager cwd, not the root); mounts (///rel) survive + render the right tree. Also the ruling's case (2) "absolute REPLACES" is not honoured by the current _runSpell (renders in-context, ctx unmoved) — separate bro-side inspection owns closing both.

TODOs

• red pty (scratchpad repro.py) from non-root PIN-1: mount click renders nothing, root click shows the launch-wt status — pinned
• tree rows mint ///-abs-context O; wt key/± U→O; check.js + getctx pins flipped to O (check.js flips RED on a base --tlv)
• green: FULL test/work/view/run.sh PASS (plain banner intact, goldens match, check.js 303 OK); mount renders the right tree from any cwd (repro.py); todo/view + graf/parity PASS

Blockers and bummers

• CORRECTION: my earlier "test/work/view pre-existing env red" claim was FALSE — the lost work banner was MY regression from an unconditional TAG_O nav in the PLAIN hunk (I mis-checked base against a jab-cat-corrupted file, never a pristine clone). Fixed by btns-gating the nav O (BE-054's recorded mode); run.sh PASSes.
• test/work/getctx run.sh stays RED at base for ONE known reason: getctx.py:89 waits on the RETIRED [get] button (WORK-004) so no frame paints — a separate //WORK-010 staged fix re-pins it; NOT touched here, NOT a WORK-013 regression.
• ROOT /// collapse + absolute-not-replacing are bro-side (above).

Outcome

• Staged (put, NOT posted): views/work/work.js, test/work/view/check.js. clickpty.py reverted to baseline (TRK-5 launch anchors to root — the bug does not repro there, so no clean flip; repro.py from PIN-1 is the real-UI flip).
• Suggested commit: WORK-013: tree/wt row clicks mint context O spells