audit: JS-tabbed pages — only the initial view is rendered, and hidden panes still get contrast-scored (Lc 0 noise)

[rz1989s]

What

Auditing a hash-routed tabbed page (5 tab-panes, only one display:block at a time):

1. Only the initial tab is audited. The 4 hidden panes never render — their layout/overflow/contrast issues are invisible to the audit. For tabbed docs/SPAs the audit silently covers ~20% of the page.
2. Hidden panes still produce contrast findings. The report contains rows like span: APCA Lc 0 < 60 / Lc 2 / Lc 5 — consistent with measuring text in display:none subtrees against the wrong background (Lc ~0 = fg/bg compute to the same lightness, or measurement of unpainted elements). This is noise that buries real findings.

Repro

node engine/audit.mjs <page-with-js-tabs>.html --out . — page uses .tab-pane{display:none} .tab-pane.active{display:block} + hash routing.

Suggestions

• Skip contrast checks on elements with a display:none ancestor (visibility-aware traversal)
• View coverage for tabbed pages, pick any:
  • accept #fragment in path mode (load file://…#flow2 so hash-routed pages work)
  • --view-script <path> executed before each screenshot pass
  • auto-detect [data-tab-view] / .tab-pane targets and render one screenshot set per view, reporting per-view gates
• Report should state which views were audited (coverage honesty — analogous to test coverage)

Environment

keystone @getpipher/keystone (pi npm install), engine/audit.mjs path mode, 2026-08-30 run on a 5-tab architecture diagram.