# Focused live evidence C/E — issue 19442

Head: 3bb66f26c9baefc2541974ce8809eb30c45cbc1d. No source edits, commits, push, or forge writes.

C: Paired actual Vite web client localhost:6720 to isolated server14760 using normal one-time pairing. Seeded actual client store with exported persistClientSettingsPatch, 30 then60 to ensure normal effect rerun. Wrapped WebSocket.send only to observe outgoing requests, not replace report delivery. CE-web-report.json records normal BrowserProfileReporter preview.reportProfiles requests. Runtime agent on existing test thread used normal MCP preview tools; t3_preview_list later confirmed runtime=server. Requested800x450 CSS viewport; encoded1600x900 due server context DPR2. C-web60.mp4:90 frames/4.545967s, about19.80fps. This is a configured60 capture, not delivered60fps claim.

E disconnect: closed own app preview tab via normal t3_preview_close after parent confirmed its old app tab closed. Desktop client is on separate14761. Normal thread CLI (--base-dir .t3) sent agent request without BrowserProfileReporter. Existing server-owned fixture tab started a NEW recording. E-disconnect60.mp4:113 frames/4.321800s, about26.15fps.

Temporary Node inspector diagnostic attached to confirmed port14760 owner PID2055976, whose cwd was worktree/apps/server. A single breakpoint at actual encoder.evaluate(recordingEncoderScript(recordingFrameRate)) read closure value60 at23:31:53.876Z, then resumed immediately and disabled debugger. Agent start returned23:31:54.200Z and exported at23:31:58.375Z. This is actual runtime rate selection, not a source assertion. Breakpoint briefly paused startup; recording began after auto-resume. No permanent code changes. Extra1second diagnostic capture was unnecessarily queued before success was noticed; its transcript is retained but not needed for proof.

E restart/default30: parent stopped owned dev session68999 and restarted at23:33:21.543Z (restarted-dev.log), same isolated home/ports. Confirmed new port14760 owner PID2075231 cwd worktree/apps/server, compared old PID2055976. No web clients opened. One-shot breakpoint read actual recordingFrameRate=30 at23:34:09.116Z; auto-resumed and disabled debugger. Agent NEW server-owned tab start23:34:09.486Z, stop/export23:34:13.648Z. E-restart30.mp4:118 frames/4.343233s (~27.17fps), H.2641600x900,68,443bytes. E live lifecycle PASS. C alternate client/runtime path PASS; normal60 report plus actual server-owned artifact. No claim of achieved60fps or new smoothness proof.

Scene: reused parent fixture http://127.0.0.1:19443/ (white page with changing frame counter), not the committed moving.html scene. Appropriate for rate routing/lifecycle proof; these recordings are not a matched performance comparison.

Cleanup: no new dev/fixture servers or desktop processes started. Own app preview tab closed. Runtime-agent fixture tab remains available on restarted parent server. Temporary inspector debugging disabled after each observation; inspector listener will be closed before handoff. CLI requests completed; diagnostic helper processes have bounded90second lifetime. Parent owns restarted dev runner, desktop and fixture servers.

FRICTION: tool — startup pairing token already consumed; recovered using normal CLI pair --base-dir. Infra — server-owned DPR2 produces1600x900 from800x450 CSS viewport; delivered rates remain below requested ceiling.