feat(benchmark): scaffold Code→UI Benchmark comparators (chrome-devtools-mcp + WebMCP reference) and /benchmark v1 results page

[Narrator]

Summary

This PR delivers sprint 2613 Task B (benchmark-comparators-and-page) — the comparator implementations and the public results-page scaffold for the Code→UI Benchmark v1.

Three things land:

1. packages/domscribe-benchmark-comparators/ — new sibling package containing the Comparator interface, the WebMCP-conformant in-repo reference server (per RFC 0004's fallback — no public WebMCP-conformant server installs cleanly in-sprint), and the chrome-devtools-mcp comparator shim. 17 unit tests, 84% line coverage. The runner in @domscribe/benchmark (Task A) imports Comparator implementations from here.

2. chrome-devtools-mcp@1.1.1 pinned as root dev-dep — exact-version pin (no caret) so the published 45-cell results table is reproducible byte-for-byte. Lockfile diff includes pnpm's routine forward-resolution of cookie / cors / qs / http-errors transitive patches.

3. /benchmark static results page (docs/code-to-ui-benchmark/v1-results.md) — public artifact RFC 0004's category-naming falsifier reads. Schema-true placeholder cells; Task A's runner populates them from results.json. Linked from README.md (Features section) and TECHNICAL_SPEC.md (§9 Framework Support Matrix). Methodology, comparator table with external-validity flag, reproduction command, and falsifier text all included.

Plus: docs/code-to-ui-benchmark/vendor-submission-amendment-plan.md — the checklist for amending Sprint 2612 IDE-vendor-directory submissions with the /benchmark link while review windows are still open.

Why a separate package boundary

The comparators have a different stability surface than the benchmark spec: the WebMCP reference is intentionally drop-in replaceable once a real external WebMCP-conformant server is installable, and the chrome-devtools-mcp shim is a thin wrapper over a pinned external dev-dep. Co-locating them with the spec/runner in @domscribe/benchmark would couple two distinct evolution rhythms. Each package can publish independently when its surface changes.

Acceptance criteria mapping (PM task B)

Criterion	This PR	Notes
chrome-devtools-mcp installed as a pinned dev-dep	✅	package.json line; lockfile updated
Comparator wired to run against 5 scenarios × 3 fixtures = 15 cells	⚠️ Partial	Comparator class is wired; fixture instrumentation and the runner invocation live in Task A's @domscribe/benchmark — the 15 cells materialise once Task A's runner imports createChromeDevtoolsMcpComparator and calls run(prompt) against the fixtures
WebMCP-conformant reference wired	✅	In-repo reference per RFC 0004 fallback; /benchmark page surfaces the external-validity: in-repo-fallback caveat verbatim
/benchmark page live, linked from README.md and TECHNICAL_SPEC.md	✅	docs/code-to-ui-benchmark/v1-results.md with last-run timestamp slot, comparator versions, methodology link to the spec, and reproduction command snippet
Sprint 2612 vendor-directory submissions amended	⚠️ Conditional	docs/submissions.md and the four submission PRs are not yet on main as of sprint 2613 kickoff — see vendor-submission-amendment-plan.md for the trigger-conditional procedure
Engineer logs ≥8h of OSS reference-integration pairing time	⚠️ Out of scope for this PR	Pairing time will appear in sprint retro notes per PM acceptance

Dependencies on Task A (Sprint 2613 batch A)

This PR is structurally a Task B deliverable; Task A (benchmark-rcp-foundation — @domscribe/benchmark package, docs/code-to-ui-benchmark/v1.md spec, RCP-v1 column with 15 cells) has not yet landed on main as of this PR. The Comparator interface here was designed to be the contract Task A's runner consumes, but:

• The benchmark spec doc (docs/code-to-ui-benchmark/v1.md) — referenced from this PR's v1-results.md and from the comparators README — does not yet exist. Links resolve to a 404 until Task A merges.
• The fixture instrumentation for S1–S5 lives with Task A. Until that lands, the 15 cells per comparator column remain as — placeholders on /benchmark.

When Task A merges, the runner's import statement should look approximately like:

import {
  createChromeDevtoolsMcpComparator,
  createWebMcpReferenceComparator,
} from '@domscribe/benchmark-comparators';

No rebase conflict expected — Task A and this PR touch disjoint paths (packages/domscribe-benchmark/* vs. packages/domscribe-benchmark-comparators/*).

Pre-existing blocker (escalation, not a blocker for this PR)

Sprint 2492's RCP-v1 stamp (PE RFC 0003 falsifier) has not landed on main as of sprint 2613 kickoff — no docs/rcp/v1.md, no docs/submissions.md. The PM plan's replanning trigger #1 names this as a "full Sprint 2613 replan" trigger because the benchmark has no v1 subject without RCP v1. This PR does not depend on RCP v1 landing because the comparators are framework-agnostic, but Task A and the /benchmark page's RCP-v1 column do depend on it. Surfacing here for visibility; a separate Slack escalation to Kaushik covers the broader sprint-thesis risk.

Test plan

• pnpm nx run domscribe-benchmark-comparators:lint — passes
• pnpm nx run domscribe-benchmark-comparators:test — 17/17 tests pass; 84% statements / 71% branches / 92% functions / 83% lines coverage (all above configured thresholds)
• pnpm nx run domscribe-benchmark-comparators:build — compiles cleanly
• pnpm nx affected -t lint --base=main~1 — comparators package + all other affected packages pass; pre-existing domscribe-test-fixtures lint errors are unrelated to this PR (verified by stashing changes and re-running lint on base)
• Manual verification that /benchmark v1-results page renders correctly on GitHub — to be confirmed once Task A populates results.json
• Comparator integration test against fixtures — owned by Task A's runner PR

[nx-cloud]

View your CI Pipeline Execution ↗ for commit 9250faf

Command	Status	Duration	Result
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	1m 37s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	1m 39s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	1m 38s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	1m 20s	View ↗
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	51s	View ↗
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	43s	View ↗
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	22s	View ↗
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	15s	View ↗
Additional runs (18)	✅ Succeeded	...	View ↗

💡 Verify your cache is correct by running tasks in a sandbox. Read docs ↗

---

☁️ Nx Cloud last updated this comment at 2026-06-06 20:16:53 UTC

[Narrator]

Closing agent-generated PR during workspace cleanup.