The three current downgrade-sublibraries failures are unrelated to this generic-group fallback change. Jobs 87326748367 (OrdinaryDiffEqCore), 87326748413 (OrdinaryDiffEqDifferentiation), and 87326748673 (StochasticDiffEq) all fail resolution because their explicit SciMLOperators floor is 1.23.0 while current OrdinaryDiffEqNonlinearSolve requires 1.24.3. Draft PR #3905 contains the focused floor correction and exact strict-floor local verification for these projects.

Updated failure classification after more cells completed:

• The four SciMLOperators 1.23.0 downgrade resolver conflicts (OrdinaryDiffEqCore, OrdinaryDiffEqDifferentiation, OrdinaryDiffEqFIRK, and StochasticDiffEq) are the exact metadata issue fixed and locally verified in Raise SciMLOperators downgrade floors to 1.24.3 #3905. This branch intentionally does not duplicate that commit.
• The BDF and OrdinaryDiffEqNonlinearSolve downgrade failures are also present on Raise SciMLOperators downgrade floors to 1.24.3 #3905 itself (BDF, NonlinearSolve); they are not caused by the four-line GROUP forwarding change here.
• DelayDiffEq [QA] now runs rather than silently falling back to ALL, and reports 16 passing QA checks followed by two genuine ExplicitImports errors. The same failure exists on Raise SciMLOperators downgrade floors to 1.24.3 #3905 (job). A clean-base reproduction/public-API investigation is queued separately per the repository instructions.
• DiffEqBase [Downstream2] executes the community callback suite and reaches the known obsolete callback assertion; a clean-base reproduction/bisect is already in progress.

Thus the newly visible QA/downstream failures are substantive tests exposed by the selector repair, not evidence of another silent no-op.

The clean-base DiffEqBase Downstream2 callback failure exposed here has now been bisected and fixed in SciML/OrdinaryDiffEq.jl#3914. The first bad SciMLBase migration commit was ba118add; the fix restores the old upcrossing-only event behavior with the current three-argument VCC constructor. Local Julia 1.12 verification passed the targeted solve and the full Downstream2 group (Community 16/16, Distributed 4/4, package exit 0).

CI timer check (2026-07-15 09:49 EDT): the changed downstream Integration workflow run (29405724542) is terminal with 12/12 jobs successful. This confirms that the generic group values introduced by this PR are accepted and those downstream jobs actually execute.

The repository-wide rollup is not terminal yet: 152 checks have succeeded, 12 have failed in unrelated monorepo downgrade/sublibrary suites, 3 were skipped, and 5 remain queued (four long StochasticDiffEq groups plus the GPU job). The completed failures are being reproduced separately on a clean master checkout; they are outside this workflow-only diff. The branch is still current with upstream/master (git log HEAD..upstream/master --oneline is empty).

⎿ Setting 3-hour timer for CI completion as per CLAUDE.md instructions...
Current time: 2026-07-15 09:49 EDT
Will check CI results at: 2026-07-15 12:49 EDT

Re-audited the current registered-package downstream workflow inventory on 2026-07-26. There are 23 active shared-workflow lanes targeting the four migrated runners covered here: DiffEqBase (12), DelayDiffEq (5), StochasticDiffEq (5), and DiffEqDevTools (1). Each currently exports only generic GROUP, while the default-branch runners read only ODEDIFFEQ_TEST_GROUP; their requested lanes are therefore ignored.

Fresh local verification from this PR head: julialauncher +1.12 --startup-file=no test/qa/sublibrary_group_selection.jl => 16/16 passed. This confirms generic GROUP selection, package-specific override precedence, and each original default.

Rebased onto current master (ffda967d1) and force-pushed with lease: 96008a42.

Local verification after rebase:

• julia +1.12 --startup-file=no test/qa/sublibrary_group_selection.jl passed: 16/16.
• git diff --check upstream/master...HEAD passed.

Rebased onto current master (8037f9f6, including merged #3904) and force-pushed with lease: 7379cd7f.

Local verification after rebase:

• julia +1.12 --startup-file=no test/qa/sublibrary_group_selection.jl passed: 16/16.
• git diff --check upstream/master...HEAD passed.

The prior CI’s seven red lanes all reached the test runner (one was canceled while in it); they are live test/infrastructure outcomes, not group-routing silence.