Side-by-side editor demos should support single-engine mount for honest metrics

Problem

A side-by-side editor demo is fine for eyeballing behavior, but it is a bad default measurement surface. Mounting both editors at once adds extra DOM, layout, and page-level work that can make one engine look slower than it is.

The Huge Document docs page was doing exactly that: it rendered Plate and Slate next to each other, then showed timing stats as if those numbers were clean per-engine measurements.

Symptoms

• The docs page looked like a comparison harness, but it mounted two huge editors at once.
• Temporary automation against the page showed the Slate column in Plate + Slate mode reporting a visible slowdown in one run:
  • last keypress: 10 ms
  • average of last 10 keypresses: 21 ms
  • last long animation frame: 73 ms
• The same page in Slate only mode dropped back to the near-zero / non-reporting class, which is where standalone Slate also tended to land under automation.

What Didn't Work

Treating the side-by-side view as both demo and benchmark surface was the mistake. The numbers looked comparable, but the page itself was part of the overhead.

Copying the local .tmp/slate-v2 transform-benchmark controls into the docs page would have made this worse. That work belongs in a dedicated benchmark harness, not in a public example page.

Solution

Add a Mounted editors control with:

• Plate + Slate
• Plate only
• Slate only

Keep Plate + Slate as the parity/demo view, but make single-engine mount the obvious truth mode for anyone looking at timing stats.

Also add an explicit note in the stats panel:

Mount one editor at a time for cleaner engine-specific numbers. The two-editor view is useful for eyeballing parity, not for honest perf baselines.

Why This Works

The control separates two jobs that were previously mixed together:

• demo job: show both editors on the same page
• measurement job: remove cross-engine page overhead

That keeps the docs page useful for both audiences without pretending the two-editor layout is a neutral measurement surface.

Prevention

• If a docs page mounts multiple heavy editors, assume it is a contaminated timing surface until proven otherwise.
• Use dedicated benchmark routes like /dev/editor-perf for transform and mount investigations.
• If a comparison page still shows UI metrics, give it a one-engine-at-a-time mode.