[Malena-yy]

Summary

• Give User Memory and Agent Memory guides a common reading path, with matching
  English/Chinese heading structures and command examples. Keep track-specific
  benchmark and lifecycle details rather than forcing identical content.
• Separate bounded smoke checks from full evaluations; consolidate repeated
  commands and link to the existing parameter, dataset, and test references.
• Maintain numeric result tables in the English result pages only. Add a User
  Memory Chinese results landing page and make the Agent Memory Chinese page a
  translated explanation with direct links to the canonical tables.
• Clarify PersonaMem v2 option-matching versus LLM judging, plugin lifecycle
  data risks, and the distinction between resuming stages and configuration replay.
  Correct replay argument order and quote cleanup placeholders.
• Standardize navigation, track names, and Chinese architecture terminology.
  Retain all previous heading anchors on changed pages as compatibility aliases.
• Remove unexplained star/underline/bold score decorations in Agent results and
  label accuracy units explicitly. Do not infer ranking or significance claims.
• Document the bilingual/single-source maintenance convention in CONTRIBUTING.

Validation

• python -m pytest -q: 198 passed, 20 subtests passed in the existing isolated
  Python 3.11 development environment.
• python scripts/check_doc_links.py and git diff --check: passed.
• A one-off local documentation audit additionally verified:
  • Matching English/Chinese guide heading levels and Bash/YAML command blocks.
  • All existing User Memory result-table lines and all 160 Agent Memory
    numeric result cells unchanged; the previous Chinese Agent table had the
    same numeric cells as the English canonical table.
  • All repository Markdown file targets and 44 local fragment links resolve,
    including the new untracked-at-audit-time Chinese results page.
  • Markdown headings render correctly and all previous heading anchors on
    changed pages remain available.
  • 24 Bash example blocks parse, and documented flags match the help output
    for 8 runners (using the Python entrypoint help for AgentBench).
• Verified that the newly linked upstream EvoAgentBench GitHub project exists.

No runtime code, benchmark scores, dependency lists, or workflow settings changed.
No live benchmark, paid model call, installer, or memory deletion was executed.