client: at the factory row pitch a forward wire is drawn through the cards of the next row (#2161 regression) #2164

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: a6dd7dd7e74b4328a9bec47d8a8ce344
Branch: dev/2-forward-wire-factory-pitch
Base: 96e24ef
At: 2026-11-02T12:55:38Z

What goes wrong

#2161 (merged as 96e24ef307e034f3be2706ac60a6631cc8db0e24, 12:54:38Z) made a row-crossing wire measure its corridor from the row's lowest measured bottom. In the layout the app actually produces, a row's cards can be taller than the row pitch, so that bottom is already below the next row's top: the corridor gets clamped to zero and lands inside the cards of the row below.

Concretely, with the factory pitch (plugins/workflow/src/graph.js:10 GAP_Y = 108, layoutNodes places a second row at y = 108):

Edge a → c:

• at 0bfaf525 (before #2161): M 0 48 L 0 108 — straight down, clean;
• at 96e24ef3 (now): M 0 48 L 0 120 L -96 120 L -96 108 L 0 108 — the run at y = 120 sits inside c (its body is 108–180), and the final approach rides c's top border.

Same layout, edge a → d: one card crossed before (b, which is what #2161 set out to fix), three now (b, c, d). Heights of 140 and 160 behave the same; at 108 the corridor collapses onto the target's top border.

Cause: rowBottom = max(bottom of boxes whose top is within 1.5px of the source's top) is 120 here, while to.top is 108, so gap = min(WIRE_GAP, max(0, (to.top − rowBottom) / 2)) is 0 and yA = rowBottom. #2161's two tests place the second row at y = 120 / y = 400 (a 280 pitch), so they never exercise a row that starts above the previous row's tallest bottom.

Reproduction

Build the graph above with normalizeGraph / the page's own layout, then read the wire's d for a → c: it contains a horizontal run at y = 120, inside c. Same with h = 140 / 160; h = 108 puts the run on c's top border.

Expected

• A forward wire never enters or rides the border of a non-endpoint card in either row, including when the next row begins above the source row's tallest bottom.
• The #2161 case still works: with a 96–120px card beside a 48px source and a row far enough below (the 280 pitch used by its tests), the wire leaves above the neighbourhood and does not cross the neighbour.
• No layout that was clean before #2161 becomes dirty: for every graph, the number of cards a wire crosses must not increase relative to 0bfaf525.
• The same-row (loop/skip) and back-edge geometries keep their current d strings verbatim (#2151 10 cases, #2138 6 cases).

Suggested fix shape

When the row's lowest bottom leaves no room for a corridor — to.top − rowBottom < 2 × WIRE_GAP — fall back to the pre-#2161 corridor (source-only, clamped to stay above to.top), so factory-pitch graphs render exactly as they did before; keep the row-measured corridor only when there is room. The guard must also make yA > yB unrepresentable.

Scope

core/src/client/workflows.js, core/test/workflows-page.test.js (one test on the factory 188/108 pitch with a 120px card, not a hand-picked 280 pitch), core/client.js + patches/ds-bot/0008-client-bundle.patch + the two patches/ds-bot/files/ mirrors via build + generate, docs/subsystems/core-client.md sentence, changelog fragment.

Duplicate check

row pitch wire (0), GAP_Y (2: this regression's own PR #2161, closed, and an unrelated panel issue #1513), "next row" wire in titles (0). No other issue covers the factory-pitch collapse; this is a follow-up to #2160, not a duplicate of it.