Beagle SCM

TODO-011: split the one-ticket page out of todo into its own view ticket

RULING (gritzko, 2026-08-04): todo is the TABULAR side only — the open-ticket board, a topic's list, the meta-pair filters. The one-ticket page becomes its OWN view named ticket. Today one view multiplexes both by arg shape (views/todo/todo.js:1-6: bare → board, ABC → topic list, ABC-123 → ticket page) and has grown to 1601 lines; the page arm and the table arm share an arg grammar they don't actually share (filters mean nothing on a page, a page key means nothing to a table).

Input

Context

• The routing is LEXICAL (the uc ucnum* "-" dgt+ key rule, todo.js:4-5) and stays lexical — no path resolution in arg classing.
• The page arm renders a thin todo/GET/GET-001.mkd or a fat todo/GET/GET-001/README.mkd off be.todoRoot() (URI-016 climb, never an env var).
• Click spells are the coupling to cut carefully (BE-054: context-less O spells, the pager stays arg-blind): 1. board/list ticket-key rows carry todo <KEY> spells today — a KEY row must now spell ticket <KEY>. 2. TODO-008 in-page ticket-valued meta links (See: BE-050) navigate to the ticket page — now ticket BE-050. 3. filter/topic spells (todo ABC Now:OPEN, argLineWith replacement, BRO-025) are table business and STAY todo spells, untouched.
• Meta-pair machinery (TODO-003 metaidx, ONE find() per verb run; TODO-004 filters; the Now:/Sev: reading with the legacy header-mark fallback) is shared ground: the page shows a ticket's own pairs, the table filters on them. Share via requires, never duplicate.
• Siblings: TODO-004 (arg grammar), TODO-008 (ticket-valued links), TODO-005 (board row growth, open).

Goals

ticket ABC-123 renders the one-ticket page exactly as todo ABC-123 does today (byte-parity is the RED/GREEN baseline); todo keeps board + topic list + filters only.

• todo ABC-123 no longer serves the page: it refuses in plain words pointing at ticket ABC-123 (errors-in-plain-words rule; if a silent re-route is preferred that is gritzko's call — record the question in the Outcome, default to the refusal).
• Every ticket-key click anywhere in the two views lands in ticket; every filter/topic click stays in todo.
• ticket rejects non-key args in plain words (ticket ABC names the topic list it is not: point at todo ABC; filters likewise point at todo).

Constraints

• Verbs resolve args, the pager stays arg-blind; spells stay context-less O spells (BE-054); arg classes stay LEXICAL.
• Shared helpers (metaidx read, mkd page rendering, key lexing) move to a required module or stay required from one home — NO second implementation of any of them.
• todo/done/ never lists — unchanged; the ticket view CAN render a closed ticket by key (a page is not a listing; same fail-open spirit as todo.js:45-48).
• Minimal moving: extract, don't rewrite; views/ticket/ + edits in views/todo/; registry entry for the new view; no other verbs.
• Status/board output for the table arms stays byte-identical; the page under ticket stays byte-identical to today's page under todo (modulo the in-page spell verb name).

WIP

Design decisions

• *No registry entry exists to add.* core/registry.js build() resolves a bareword by FILE (views/<verb>/<verb>.js, then verbs/<verb>/) — dropping views/ticket/ticket.js in registers the view. Same for the test suite: test/CMakeLists.txt globs <verb>/<case>/run.sh, so test/ticket/page/ registers itself as be-js-ticket-page.
• *ONE home for the shared ground, and it is views/todo/todo.js.* The page arm MOVED to ticket.js (refdefs, targetSpell, emitPage, pageLinks, the two tok helpers — 135 lines, cut verbatim); everything both views need stays in todo.js and is required from there: the key/topic LEXER (shape, ticketKey, keyTopic), the board root (boardDir), the page-file ladder (pageFile), the byte read (readBytes), EXTS, the arg-line grammar (parseArgs, argLineWith, spellWith) and the two click classifiers (navSpell, filterVal). Nothing is implemented twice; the edge is one-way (ticket.js → todo.js), so it needs no lazy require (unlike work.js ⇄ todo.js).
• *parseArgs stays purely GRAMMATICAL and is shared.* It classes a token lexically (uc ucnum* "-" dgt+ unchanged) and never decides which view serves what. Its old "a ticket id takes no filters" check MOVED out to the two verbs, so each refusal can point at the other view and neither message is duplicated. unscheme grew a scheme parameter so the ONE DIS-060 parse serves todo:GET-1 and ticket:GET-1 alike.
• *todo ABC-123 REFUSES (the ticket's stated default), it does not re-route.* The arg line IS the address bar (BRO-024/BRO-025), so a todo line answering with a page would lie about where the reader is. One refusal covers the filtered form too.
• *Which spell is which:* a spell that lands on a PAGE is a ticket spell; a spell that lands on a LISTING is a todo spell. That is the whole rule, and navSpell() is its single point of truth for ticket-valued links. The in-page meta KEY/VALUE filter halves therefore stay todo spells minted through spellWith — a cross-view landing BY DESIGN, since a page carries no filters (a ticket-id arg line resolves to its TOPIC via argLineWith).
• *views/work/work.js had to move too* (one literal): the [?] invite on a ticket-named worktree minted //: todo <KEY>, which the refusal would have turned into a dead button. It is //: ticket <KEY> now; the TOPIC-named branch is a listing and stays //: todo <TOPIC>.
• *core/loop.js gains wrap.ticket = false.* An UNLISTED view type defaults to SOFT-wrap (pager.js wrapFor), so without the seed the split would have silently re-wrapped every ticket page. false preserves exactly what the page had under todo.
• *The ticket miss is PLAIN WORDS, not the board's legacy TODONONE.* todo's own miss line is golden-pinned (BE-038) and is untouched; the new view starts clean under the errors-in-plain-words ruling.

TODOs

• Baseline capture: today's todo ABC-123 page bytes for a thin and a fat ticket → the ticket parity fixture.
• views/ticket/ticket.js extracted; registry entry (none needed — resolution is by file); todo page arm removed with the plain-words pointer.
• Spell sweep: key rows + TODO-008 value links → ticket <KEY>; filter/topic spells untouched; pager click round-trip verified.
• Tests: page parity (thin + fat), the todo ABC-123 refusal, ticket non-key refusals, existing todo suite green.

Blockers and bummers

• The suite run overlapped another worker's ctest on the same box; the first pass showed ~15 unrelated 10 s-TIMEOUT rows (post/patch/diff cases) that pass on a re-run. be-js-todo-rowcache was ALREADY red on the base (TODO-010) and is not touched here.

Outcome

DONE — views/ticket/ticket.js is the one-ticket page view; todo is the board, the topic lists and the meta-pair filters, and nothing else.

Baseline parity (the RED/GREEN evidence)

The pre-split tree (be @ 94714be8, untouched copy) and the post-split tree were run over the SAME fixture — a thin todo/PAG/PAG-001.mkd and a fat todo/PAG/PAG-002/README.mkd, each carrying meta pairs (plain, ticket-valued, an UNREGISTERED Zzz:, a colon-carrying Rev:), a bare key mention, a [KEY] reflink, a named refdef reflink and a /wiki/Sample shortcut:

• --plain: todo KEY vs ticket KEY differ in ONE byte run, the banner (todo PAG-001 → ticket PAG-001). Both goldens are committed under test/ticket/page/golden/.
• --tlv: the VISIBLE token stream (every non-O token, tag + bytes) is byte-identical, 111 tokens on each page.
• --tlv spells: the ONLY delta is the verb name on the spells that land on a PAGE.

- SPELL " PAG-002" -> "todo PAG-002"        (See: value)     + "ticket PAG-002"
- SPELL " PAG-003" -> "todo PAG-003"        (Zzz: value)     + "ticket PAG-003"
- SPELL "PAG-002"  -> "todo PAG-002"        (bare body key)  + "ticket PAG-002"
- SPELL "[/PAG-003/]" -> "todo PAG-003"     (reflink run)    + "ticket PAG-003"
SPELL "Now:"     -> "todo PAG Now:*"      UNCHANGED (filter → the board)
SPELL " OPEN"    -> "todo PAG Now:OPEN"   UNCHANGED
SPELL "Sev:"/" MED"/"See:"/"Zzz:"/"Rev:"  UNCHANGED
SPELL "W"/"Sample" -> "cat wiki/Sample.mkd"  UNCHANGED

The board and topic listings show the mirror image: every KEY row went todo <KEY> → ticket <KEY>, while the topic header (GET → todo GET) and the [ ✔]/[ ✗] panel spells (done KEY / dont KEY) are untouched, and the visible tokens are byte-identical (41 board / 37 topic).

Refusal texts (all plain words, no bare code)

$ todo PAG-001
todo: 'PAG-001' names one ticket page, and this is the board — write 'ticket PAG-001'

$ todo PAG-001 Now:OPEN todo: 'PAG-001' names one ticket page, and this is the board — write 'ticket PAG-001'

$ ticket ticket: which ticket? — write 'ticket ABC-123' for one ticket's page, or 'todo' for the open board $ ticket PAG ticket: 'PAG' names a topic, and a topic is a LIST — write 'todo PAG' for its open tickets,

$ ticket Now:OPEN ticket: 'Now:OPEN' is a filter, and a filter narrows a LISTING, never a page — write

$ ticket PAG-001 Now:OPEN ticket: 'Now:OPEN' is a filter, … — write 'todo PAG Now:OPEN' for that list; … $ ticket PAG-999 ticket: there is no ticket PAG-999 under todo/PAG/ — 'todo PAG' lists that topic's open tickets

`todo`'s own topic miss keeps its golden-pinned `todo: NOPE: TODONONE` line; every exit is non-zero.

Spell sweep, through the REAL pager (pty clicks, be-js-todo-nav)

ok  page-painted                        ok  board-painted
ok  see-value-navigates                 ok  board-inline-value-navigates
ok  see-value-is-not-a-filter           ok  board-inline-value-is-not-a-filter
ok  unknown-key-value-navigates         ok  list-painted
ok  key-half-still-filters              ok  list-key-row-opens-the-ticket-view
ok  key-half-answers

Each landing is asserted on BOTH the BRO-024 address bar (/: ticket NAV-002) and the painted body, so a todo answer that merely mentions the code cannot pass. New leg list-key-row-opens-the-ticket-view: a pty click on a topic-list KEY row lands on that ticket's page in the ticket view. be-js-todo-click pins the other direction: an in-page meta pair click still opens todo CLK Now:* / todo CLK Now:OPEN. be-js-work-view pins the work board's [?] → //: ticket PIN-1 (its pty leg fires the button and lands on the page).

Suite

• base 94714be8, before: *352/353*, one red — be-js-todo-rowcache (pre-existing, TODO-010: the and-the-row-grew-the-staged-tick pty leg).
• after: *353/354*, same one red, same assertion. The new case is be-js-ticket-page.
• (an intermediate run showed ~15 10 s-TIMEOUT rows; that was another worker's ctest on the same box, all green on the clean re-run.)

Files

NEW  views/ticket/ticket.js          260   the page view (135 lines cut verbatim out of todo.js)
NEW  test/ticket/page/run.sh         ~150  parity vs the pre-split goldens + both refusal families
NEW  test/ticket/page/check.js       ~115  the page spell sweep
NEW  test/ticket/page/golden/PAG-001.plain, PAG-002.plain

Open questions for gritzko

1. *Refusal vs silent re-route* — took the ticket's stated DEFAULT (refuse). Say the word and todo ABC-123 can forward to ticket ABC-123 instead; the refusal is one bad() call in todo().
2. *ticket's wrap default* — set to false (no-wrap) purely to preserve what the page had while todo served it. A ticket page is PROSE, so true (soft-wrap, the cat/diff/status default) may be what you actually want now that it has its own type.
3. *The ticket miss line is plain words*, not the board's legacy todo: KEY: TODONONE. The errors-in-plain-words ruling says so; the BE-038 golden keeps todo's own line as it was. Say if you want the new view to carry the old shape for symmetry.
4. *views/work/work.js was touched* (one literal) though the ticket scoped the change to views/ticket/ + views/todo/ — its [?] invite on a ticket-named worktree would otherwise have become a dead button.

Suggested commit message

TODO-011: split the one-ticket page out of `todo` into the view `ticket`