[MendixMau]

Fixes #471

Direct lane (changes to skills/, bug-logs/, bin/, project-bin/, project-tests/, pipelines)

What changed and why (one paragraph):
project-bin/page-fidelity.js contentOf() picks the wireframe's content boundary from a preference list (<main>, wf-screen, wf-shell, mockup-frame, div.main, div.content, <body>). A popup wireframe draws the caller's page behind a scrim: .popup-stage > .shell-behind + .dialog-scrim > .dialog. The shell behind carries its own div.main, so the popup was scored against the caller's page head, <h1> and card. The issue measured 29% on a popup whose fields and bindings all matched, and 13% on a throwaway wireframe. The file's own header already said popup content is wrapped in .dialog, but nothing handled it. The fix switches to the dialog boundary only with popup evidence: a .popup-stage or .dialog-scrim in the wireframe, or a page MDL whose Layout: names a popup layout (the same test the scorer already uses for its chrome set). With evidence, it scores the element whose class list contains the word dialog (dialog-head/dialog-foot never count; div/section/dialog/aside tags are accepted). If there is no .dialog, it scores the stage minus .shell-behind. Without evidence nothing changes. .dialog is the design system's own component, and a full-page wireframe may draw an open confirm with it; the first cut of this PR switched on any .dialog, which scored a grouped overview plus an inline confirm as the confirm alone (39% on master, 5% on that cut). A bare-.dialog popup wireframe with no stage or scrim still reads <body>, as before, unless its page uses a popup layout. The annotation stripping after the boundary (wf-*, anno, bind tables, .bound) is unchanged.

Golden input: tests/wave2/fixtures/page-fidelity-popup/product-newedit-describe-v023.mdl is a verbatim describe page of a real popup page (Layout: Atlas_Core.PopupLayout), mxcli v0.23.0. It is copied unchanged from the context-pack fixture's capture. The wireframe product-newedit-popup.html is hand-drawn in the issue's quoted shape, because the issue quotes the structure, not the file. CAPTURE.md beside it says so and asks for a re-capture when a real popup wireframe can be shared.

Field evidence (rule 4):
Reported by jopterhorst against release/v1.0 at 2d80287: 29% on the field popup and 13% on a minimal throwaway, every miss in .shell-behind. Reproduced here on master with the fixture: 18% (headings 0/2 actions 0/1 content 1/2 classes 2/7). Every miss came from the shell: Product catalog, Active items, New item, page-head sub page-actions card card-sub. With the fix the result is 100% (headings 1/1 actions 2/2 content 2/2 classes 6/6, with dialog-head dialog-foot reported as layout chrome). Owed by the maintainer: one run on the reporter's popup wireframe.

Impact

Row	Answer
Process: human steps or gates added/removed; does anything new block; new stage artifacts	none: the instrument is not a gate. Scoring change: a wireframe with popup evidence (a .popup-stage or .dialog-scrim, or a page on a popup layout) and a .dialog is now scored on that dialog only, so its recorded text-match moves, usually up, because the caller's page leaves the denominator. A full-page wireframe that draws a .dialog without that evidence scores exactly as before; the fixture pins this at the master number.
Dependencies: shared files touched; order vs other open PRs; new tools/platform assumptions; new artifact reads + their producer	Only project-bin/page-fidelity.js. No new reads or tools. prototype-route.js callers are unaffected (test-prototype-route.sh 14/0). No ordering constraint against #469/#470/#477 beyond a trivial CHANGELOG.md line merge.
Context cost: always-loaded words per role, base → head	none: no routed or always-loaded file touched (bin/render-routing.sh --check clean).
Behaviour: A/B for agent-read text	n/a: scripts only.
Runtime	n/a. Two more regex scans of one HTML file.

• No client data anywhere in the diff (whole-word grep of added lines for the denylist terms, /Users/ and /home/user: 0 hits; LEAKGUARD_ALLOW_NO_DENYLIST=1 bin/check-no-client-data.sh clean)
• Size cap: 6 files, about 30 lines changed in the scorer; the rest is fixtures and CHANGELOG (the branch also carries a merge of origin/master)
• Test tier reached: T1. tests/wave2/test-page-fidelity-mocks.sh project-bin/page-fidelity.js gives 78/78. It has 10 new cases:
  • the shell's h1, card and button are not scored, and the dialog matches at 100%;
  • a stage without .dialog still drops the shell;
  • popup evidence with no .dialog and no stage keeps the old div.main boundary;
  • a popup layout in the MDL alone is enough evidence;
  • the review's regression case: grouped-overview plus an inline .dialog confirm scores the master number unchanged (39%, headings 1/2 actions 1/3 content 1/2 classes 1/5).
  • On master's scorer the fixture gives 8 FAIL; on this PR's first head (160d180), 1 FAIL, which is the inline-confirm regression. test-prototype-route.sh (drives page-fidelity on a file and a route) gives 14/0. bin/check-scripts.sh, bin/check-portability.sh, bin/render-routing.sh --check and bin/check-pr-discipline.sh origin/master are clean.
• For a changed instrument: MDL golden is captured. The wireframe is drawn to the issue's quoted structure, stated in CAPTURE.md. Layouts: the scorer reads its two inputs by path; project-root discovery is unchanged. Platforms: pure Node string handling.
• For a new skill: n/a
• CHANGELOG.md line appended, crediting jopterhorst
• New bug entries: n/a

Left open: a full-page wireframe that draws an inline .dialog and a .dialog-scrim, for example a full page shown with its own confirm open over a scrim, is read as a popup and scored as the confirm. That shape needs .popup-stage/.shell-behind to be scored the other way.

https://claude.ai/code/session_01VJgWP5vEoAsNsJYqCDMGNw

---

Generated by Claude Code