workflow: the drawer's Edit plane draws the parallel frame and wires from the old 168x72 card size, so the frame is clipped and wires cross cards #1693

Implementation: FlyBot (ours) — PR follows within 24h. Do not pick up without a pickup-ok comment on this issue.

Written-by: FlyBot Lead
Claimed-by: FlyBot Lead
Claim-Id: aa5f58ca1a89429ba0f03f9922dce163
Branch: dev/3-wrapped-frame-wires
Base: dcaae9b
At: 2026-10-29T13:13:24Z

What happens

In the drawer's Edit plane (wf-canvas[data-fit="drawer"]), a 3+ way parallel step is laid out by flowLayout into rows whose cards are sized node.w × node.h — 98×96 at panel width 364 — but the parallel frame and the wires still use the fixed 168×72 card metrics.

• Frame: parallelFrameStyle (plugins/workflow/client.js:652-664, called at :1170-1177) derives left/top/width/height from NODE_W and node.y + NODE_H, never from node.w/node.h. With a 3-way step at panel 364 (cards 98×96, centres 72/182/292) it returns a frame spanning x = −22..386; the drawer canvas clips it (overflow:hidden, :2202), so both uprights are cut off. The same graph in the fitted plane (old 168-wide boxes) gets a frame of 2..362, inside the panel.
• Frame bottom: at 96px cards the frame's bottom edge lands at y=318 while the cards reach 352, so the dashed bottom edge runs 34px under the cards (10px at the 72px fallback — the same arithmetic, just enlarged by the lay-out).
• Wires: the laid-out plane draws with wirePath(from, to, 0) (:1158-1163, one fold at the midpoint of the two rows). On the 6-way fixture that puts the horizontal segment through cards on 8 of 13 edges — c>r4 at y=258 through r1/r2, c>r6 through r2/r3, r1>d at y=385 through r4/r5, r3>d through r5/r6. The same graph in the fitted plane has 0 crossings. flowWires (:820-824) already documents exactly this: a straight run across a band cuts through another line's cards, which is why the Read-only plane routes around them.

This was merged with #1683 / PR #1683, whose overlap work only moved the cards; the frame and wire geometry were not part of that change.

Reproduction

• Probe: tmp/audit-1029y/probe-frame.mjs (exit 1, frame and wire numbers above) and its control tmp/audit-1029y/probe-before1683.mjs (fitted plane: frame 2..362, 0 crossings).
• In the UI: open any workflow with a 3+ way parallel step in the drawer → Edit tab → panel width 364 (3-, 4- and 6-way graphs).

Expected

• The parallel frame encloses every card of the layer and stays inside the panel, for both the 96px and the 72px card height.
• No wire crosses a card in the laid-out plane (6-way fixture: 13 edges, 0 card crossings).
• The fitted plane and the Read-only plane draw exactly what they draw today.

Scope

• plugins/workflow/client.js and its test only:
  • parallelFrameStyle uses node.w ?? NODE_W and node.h ?? NODE_H.
  • The laid-out plane draws its wires from the boxes/bands flowLayout already computes, through flowWires; the fitted plane keeps wirePath.
• Numbering/positions must come from the one place that computes them (no second copy of the row geometry).
• Not this issue: dragging a card in the laid-out plane committing panel coordinates into the stored graph — that is workflow: a drag in the drawer's laid-out Edit plane commits panel coordinates into the stored graph, so the node also moves on the Settings canvas #1686. Same file, so the two branches coordinate: rebase whichever lands second.
• verify green with ARCH_DOCS_BASE=<base>; patches/** untouched (no core/ change).

Acceptance

(a) frame left ≥ 0 and right ≤ panel width on the 3-way panel-364 fixture, and it encloses every card box.
(b) the same holds at card height 72 and 96.
(c) 6-way fixture: 0 edges draw their horizontal segment through a card box (port and mini-button areas included, not only card centres).

Reading of (c) after the fix, recorded here so the reviewer and the branch agree: the drawn geometry makes "mini-button area" impossible to clear together with (e). The mini-button box overhangs the card top by 8px and the port dot by 5px, while FLOW_WRAP_GAP is 12px, so any horizontal segment running through the wrap gap lands in one of the two boxes; the 8px overhang is exactly the routing the Read-only plane draws, which (e) freezes byte-identical. The acceptance is therefore card box + out-port dot = 0 crossings; the mini-button box is documented in the test and the PR body (wires render beneath the cards, the buttons stay clickable). Clearing it would need a larger wrap gap — a change to flowLayout, which would move the Read-only plane and break (e).
(d) #1683's six byte-identity fixtures and plugins/workflow/test/client.test.js:1145-1171 are unchanged and still pass.
(e) Read-only plane unchanged: flowWires paths on the existing fixtures are byte-identical.
(f) TZ=UTC ARCH_DOCS_BASE=<base> npm run verify exit 0, red-first test for (a)–(c) on the pre-fix source.