[simone-chen]

• Adds explicit literal campaign exclusions after inclusion filters and before shuffling or sampling. This lets the B200 refresh collect affected GEMM scopes while leaving the separately tracked GLM shape expansion out of scope.
• Reports generated, selected, outside-inclusion and explicitly excluded counts. Exclusions do not change native case objects, callbacks, checkpoint identities or failure classification, and must not be derived from failed cases.
• Applies the same selection rule to pooled and full-node collectors; documents the CLI. Fixes the full-node test fixture's dependence on another test initializing the logger.
• Validation: 14 focused selection, native-callback forwarding and real checkpoint recovery tests passed; Ruff and diff checks passed.
• Stacked on the existing framework-refresh branch for review. No merge or default promotion is requested.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[simone-chen]

/ok to test 06a91a4

[simone-chen]

Closing as outside the requested review scope: retain only attention #604, recurrent/MoE #605, and communication #591 for Tianhao's review. This campaign infrastructure change is not merged or included in the replacements. Its branch and history remain available for separate future work; collection remains stopped.