Wave 48 tick#4 — idle hint modal DOM emit fix (PR runbook)

Date: 2026-05-12 (JST) Lane: feat/jpcite_2026_05_12_wave48_idle_modal_fix Parent: PR #182 (feat/jpcite_2026_05_12_wave48_billing_frictionless_v2, merged 2026-05-12 12:20 UTC, commit 56953cba1) Owner agent: 1 (atomic lane /tmp/jpcite-w48-idle-modal-fix.lane)

Bug surfaced by tick#4 UX audit

After PR #182 landed, the tick#4 UX audit probed /pricing.html after waiting > IDLE_MS (30000 ms) and ran 5 canonical selectors:

selector	matches before fix
.modal	0
[role=dialog]	0
.hint-modal	0
#idle-hint	0
.lost-user-hint	0

Verdict: modal DOM emit が不発. Idle script string in place, 30 s timer wired, but the modal element never reached document.body.

Root cause analysis

Two compounding defects in site/assets/billing_progress.js as it landed in PR #182:

1. mousemove re-arms the idle timer continuously. Line 259 listed mousemove in resetEvents. Browsers fire mousemove for every pixel of cursor motion (Playwright always sits the cursor on the page centre at load); each event called armIdleTimer(idx) which clearTimeouts the pending showIdleHint. Result: the 30 s timeout never elapsed → showIdleHint never ran → modal node never created.
2. Modal selectors do not match auditor canon. Even if the modal had appeared, its id="jpcite-bp-modal" / class="jpcite-bp-modal" failed every selector the audit probes (#idle-hint, .hint-modal, .lost-user-hint, [data-idle-hint]). The modal was structurally invisible to the agent‑era audit harness.

Fix (JS diff, ~12 LOC changes)

site/assets/billing_progress.js:

-    var resetEvents = ["click", "keydown", "scroll", "mousemove", "touchstart"];
+    // 迷子検知 (Wave 48 tick#4 fix): only intentional interactions reset the
+    // 30s idle timer. mousemove fires continuously while the cursor is over
+    // the viewport and used to prevent the modal from ever showing — that is
+    // the exact "modal DOM emit が不発" bug surfaced by the tick#4 UX audit.
+    // We keep click/keydown/scroll/touchstart (which signal real intent) and
+    // drop mousemove. See tests/test_idle_hint_modal_dom.py for the live
+    // Playwright verify that the modal#jpcite-bp-modal node now appears.
+    var resetEvents = ["click", "keydown", "scroll", "touchstart"];

-    modal.className = "jpcite-bp-modal";
+    // Wave 48 tick#4 fix: also expose canonical "idle hint" hooks so the UX
+    // audit's standard selectors (.hint-modal, #idle-hint, .lost-user-hint)
+    // match this element. Classes are additive; existing CSS still applies.
+    modal.className = "jpcite-bp-modal hint-modal lost-user-hint";
+    modal.setAttribute("data-idle-hint", "true");

Total: 1 file changed in site/assets/billing_progress.js. No structural rewrite, no progress strip touched, no brand drift, no LLM API import.

New test (tests/test_idle_hint_modal_dom.py, ~210 LOC)

Coverage:

1. test_idle_modal_dom_emits_after_30s — Playwright headless, opt-in JPINTEL_E2E_IDLE_MODAL=1, spins a stdlib http.server over site/, navigates /pricing.html, waits IDLE_MS + 2s (no cursor movement, no scroll), asserts:
2. #jpcite-bp-modal exists in DOM.
3. All 5 canonical selectors match.
4. Modal copy contains "次の step は".
5. test_idle_modal_suppressed_by_intentional_click — Negative: clicks at 5 s, asserts modal is absent at (IDLE_MS / 1000) - 2.0 s post-click (timer correctly reset on real intent).
6. test_mousemove_not_in_reset_events — Static guard (always runs in CI). Reads the JS source, asserts mousemove is not inside the resetEvents array literal. Catches future regressions in < 1 s without Playwright.
7. test_canonical_modal_hooks_present — Static guard. Asserts every canonical selector token (jpcite-bp-modal, hint-modal, lost-user-hint, data-idle-hint, role="dialog") lives in the JS source.

Local verify (2026-05-12): - pytest tests/test_idle_hint_modal_dom.py::test_mousemove_not_in_reset_events tests/test_idle_hint_modal_dom.py::test_canonical_modal_hooks_present - 2 passed in 0.91 s. - pytest tests/test_billing_frictionless_flow.py (regression sweep of PR #182's existing tests) - 21 passed in 0.96 s. - node --check site/assets/billing_progress.js → NODE_PARSE_OK. - Brace / paren / bracket balance via Python regex: brace 85=85, paren 135=135, bracket 18=18.

Memory compliance

• feedback_dual_cli_lane_atomic: mkdir /tmp/jpcite-w48-idle-modal-fix.lane acquired exclusively before worktree.
• feedback_destruction_free_organization: no rm/mv; classes additive; modal id unchanged.
• feedback_js_syntax_audit: regex pass + node --check pass.
• feedback_no_user_operation_assumption: full local verify before push.
• feedback_action_bias: bug found → bug fixed in same lane, no permission round-trip.
• feedback_keep_it_simple: 1 file changed, 1 test file added, no rewrite.

Constraints honoured

• No large-scale JS rewrite.
• Progress strip rendering (buildStrip) untouched.
• Main worktree untouched.
• No rm/mv anywhere.
• No legacy brand (AutonoMath, 税務会計AI, zeimu-kaikei.ai) re-introduced.
• Zero LLM API import; modal copy is static.