Phase 2, Task 10: Chrome DevTools MCP baseline and comparison #577
Description
Parent: #563
Phase: #567
Depends on: Task 9
Summary
Capture before/after performance metrics using Chrome DevTools MCP to validate the streaming optimization.
Steps
Baseline capture (on main)
-
navigate_pageto publisher proxy URL via Viceroy -
list_network_requests— record TTFB, TTLB, transfer size for document request -
lighthouse_auditwith["performance"]— record TTFB, LCP, Speed Index, TBT -
performance_start_trace→ load page →performance_stop_trace→performance_analyze_insight -
take_memory_snapshot— record JS heap size - Repeat 3-5 times for stable medians
Feature branch capture
- Repeat same measurements on feature branch
Comparison
- TTFB: minimal change expected (gated by backend response time)
- TTLB: should be reduced (body streams incrementally)
- LCP/Speed Index: should improve (browser receives
<head>sooner) - Transfer size: unchanged (same content, same compression)
- Response body hash via
evaluate_script: must be identical (correctness)
Red flags
- TTFB regression → investigate
finalize_response()/ synthetic ID timing - Body mismatch → bisect between Phase 1 and Phase 2
- LCP unchanged → check if lol_html emits
<head>injection in first chunk
Plan
See docs/superpowers/plans/2026-03-25-streaming-response.md — Task 10