[bompus]

The benchmark page and README merged in #58 say five rounds for every memory figure, though s3-lite-client's round 4 ran out of attempts and 42 of its cells have four. They also say the four added clients have no speed, memory, import or size measurements, though the same page now reports them. A cross-vendor read (GPT-6.1 Sol) of the merged head found these and a few smaller ones; each was checked against the saved results before changing it.

• Rounds. The caption's round count now covers every displayed arm ("4 to 5"), and the text names the exception.
• Multipliers. The sentence that every multiplier clears twice its band now says a multiplier followed by "no clear change" is a ratio of medians only. The memory report defines ncc the same way.
• Four clients. The paragraph saying they have no measurements now points to the Speed, Memory and Size sections and says only the other-service timings leave them out.
• Earlier statements. The old "imports in about 1.0 ms" clause is removed (that figure came from the previous import-only timer), and the CPU warm-up explanation is labelled as earlier experiments, not repeated on Node 26.13.0.
• Reproduction text. A unit that used its four attempts is a recorded gap with exit 9, not exit 7; other-service timings point at the release 1.0.4 protocol.
• README. The Bun lower endpoint is 0.65 µs (it was rounded twice); the chart's alt text says nine other libraries.

No measurement, summary or chart value changed; bun run check passes.

[bompus-ai-review]

Bompus AI Review (GPT-6.1 Terra, max reasoning): 1 finding(s).

[bompus-ai-review]

Bompus AI Review (GPT-6.1 Terra, max reasoning): 3 finding(s).

[bompus-ai-review]

Bompus AI Review (GPT-6.1 Terra, max reasoning): 0 finding(s).

Review 3: only high findings and above are posted from here on.