feat(website): sort benchmark result.

[samchon]

This pull request introduces improvements to the sorting logic for benchmark experiments, specifically focusing on how vendor/model strings are compared and ordered. It refactors the comparison logic into reusable functions, adds comprehensive tests for vendor comparison, and updates the main sorting call to use the new logic.

Sorting and Comparison Logic Improvements:

• Added a new compare function to AutoBeReplayComputer that sorts benchmark results first by aggregate score (descending), then by vendor/model string using the new compareVendors function. (packages/benchmark/src/replay/AutoBeReplayComputer.ts packages/benchmark/src/replay/AutoBeReplayComputer.tsL19-R45)
• Implemented the compareVendors function, which parses vendor/model strings to compare model families, parameter sizes, and versions in a meaningful order (e.g., larger parameter sizes and higher versions come first; vendors are sorted alphabetically otherwise). Also added a helper parsePart function for parsing string segments as numbers. (packages/benchmark/src/replay/AutoBeReplayComputer.ts [1] [2]
• Updated the main benchmark publishing script to use the new AutoBeReplayComputer.compare function for sorting experiments, ensuring consistent and correct ordering. (test/src/archive/publish.ts test/src/archive/publish.tsL47-R47)

Testing Enhancements:

• Added a comprehensive test suite for compareVendors, covering various real-world and edge cases to ensure correct sorting by parameter size, vendor, version, and model variant. (test/src/features/benchmark/test_benchmark_compare_vendors.ts test/src/features/benchmark/test_benchmark_compare_vendors.tsR1-R95)

Interface Update:

• Imported the IAutoBePlaygroundBenchmark interface where needed to support the new comparison logic. (packages/benchmark/src/replay/AutoBeReplayComputer.ts packages/benchmark/src/replay/AutoBeReplayComputer.tsR5)

[Copilot]

Pull request overview

This PR refines benchmark experiment sorting by centralizing comparison logic in AutoBeReplayComputer, improving vendor/model ordering semantics, and adding targeted tests for vendor comparison behavior.

Changes:

• Added AutoBeReplayComputer.compare and compareVendors to sort by aggregate score (desc) then vendor/model.
• Updated benchmark publishing script to use the shared comparator.
• Added a new test suite covering vendor/model comparison ordering rules.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 2 comments.

File	Description
packages/benchmark/src/replay/AutoBeReplayComputer.ts	Introduces shared comparators for benchmark sorting and a helper for parsing numeric parts.
test/src/archive/publish.ts	Replaces inline sort logic with AutoBeReplayComputer.compare for consistency.
test/src/features/benchmark/test_benchmark_compare_vendors.ts	Adds tests validating vendor/model sorting rules (params, variants, vendor ordering, numeric versions).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

compareVendors splits the full vendor/model slug on -, which also splits vendor names that contain hyphens (e.g. z-ai/...) and mixes vendor/model into the same token stream. This can produce incorrect vendor-level alphabetical ordering and inconsistent comparisons. Consider splitting on / first (compare vendor segment(s) separately), then apply - tokenization only to the model portion.

[Copilot]

parsePart only recognizes numeric parts that are plain numbers or ...b/a...b. Real vendor/model slugs in the repo include patterns like moonshotai/kimi-k2.5 and minimax/minimax-m2.7, where the version token has a leading letter (k2.5, m2.7). Those won’t be parsed as numbers, so compareVendors falls back to lexicographic (ascending) ordering for versions, which contradicts the intended “higher versions first” behavior. Extend parsing to handle a leading alphabetic prefix + numeric version and add coverage for these cases.

Suggested change

	const n = Number(t);
	return isNaN(n) ? null : n;
	const direct: number = Number(t);
	if (!isNaN(direct)) return direct;
	const prefixed: RegExpMatchArray | null = t.match(
	/^([A-Za-z]+)(\d+(?:\.\d+)?)$/,
	);
	if (prefixed !== null) {
	const numeric: number = Number(prefixed[2]);
	return isNaN(numeric) ? null : numeric;
	}
	return null;