[lanyue-llk]

An existing report must retain its benchmark edition when the installed task package changes. This PR carries the selected edition through the current certified runner, report builder, checker, packer and publisher.

• Verify the selected edition before provider calls, retaining the original call edition in execution metadata.
• Preserve its file and path/hash binding through ordinary builds and rebuilds. The report manifest also records the original binding so deletion or substitution is detected after rescoring selects a newer edition.
• Run the shared edition validator before every packaging early return, including frozen presentation and snapshot modes. Presentation validation uses the same check.
• Link the edition beside the home-page task scope and keep the surrounding text translatable in both languages.
• Pin the exact LexBench commit and catalog manifest hash. Only LexBench creates benchmark-vMAJOR.MINOR.PATCH tags. Browser Eval uses full commits; report rebuilds use hosting version IDs.

Validation: full local make ci passed, including 6,430 Python tests, 2,912 subtests, UI/native checks, fixture execution and installed-wheel/package checks. Boundary scenarios cover ordinary build → check → rebuild with distinct call/scoring editions, original-file tampering and binding deletion, and missing/changed sidecars, dropped bindings and wrong labels at ordinary and frozen package/publication boundaries. Existing no-edition reports remain supported.

The real 430-task, 2,580-result report passes the current checker without new calls or changed scores. Its benchmark source is benchmark-v1.0.0, with unchanged report access. The separate 550-task catalog is benchmark-v2.0.1; the patch changes version validation, not tasks or scorer rules.

Based on main after #56 merged. Companion: lexbench-webfetch #10. Main-content acceptance work remains separate in lexbench-webfetch #8.

Keep the README citation to one sentence and a BibTeX entry in both languages. Move dependency and method maintenance instructions to CONTRIBUTING, and align CITATION.cff with Apache-2.0.

[lanyue-llk]

This PR carries the selected benchmark edition through the existing runner and report pipeline so package upgrades can preserve historical measurement identities. The separate report_version module and external profile binding are a reasonable reusable design. I found no new report-instance-specific branch or copied renderer; additional object-oriented structure is unnecessary here.

Two reproduced gaps need fixing before merge: ordinary builds drop the original-call edition's execution binding, and the package/publication boundaries do not consistently enforce the new edition check. Both can be fixed in the existing shared build and validation owners.

Validation at 4221ea4: 172 relevant report, certification, dependency and packaging tests passed, 1 skipped, and 2 subtests passed. The local HTTP smoke test passed after installing the declared reference-audit dependencies. Additional negative controls reproduced both findings. In the packaging control, changing only the release version to 99.0.0 made check_release reject the report, but report_package.build and publication_check.check both succeeded. The companion lexbench-webfetch #10 also needs its scorer-input fingerprint coverage fixed.