[ChrisRackauckas-Claude]

Summary

The JET opt-mode solver paths testset in test/qa/qa.jl fails on Julia 1.13 for HCubatureJL (expects 0 reports, gets 7) and VEGAS (expects ≤2, gets more). Investigation shows this is a JET-on-1.13 analyzer bug, not an Integrals defect:

• Julia's own optimized IR for solve(prob, HCubatureJL()) is identical on 1.12.7 and 1.13.0 (code_typed(...; optimize = true)): same inferred return type, and none of the widened types JET reports (e.g. Union{Vector{Float64}, Vector{Int64}, Vector{Real}}) appear anywhere in Base's IR. The widening exists only inside JET's own re-analysis.
• JET 0.11.6 (what CI resolves) emits false-positive RuntimeDispatchReports — the same class documented upstream in False-positive runtime dispatch reports on Julia 1.13-rc1 for concrete Base broadcast/view/CartesianIndex code (clean on 1.12) aviatesk/JET.jl#839.
• JET 0.12+ crashes outright on the same analysis — typeof_arg does not support the Base.BottomRF nodes that 1.13's optimizer emits for prod/mapfoldl reductions. Filed upstream as @report_opt crashes on Julia 1.13: typeof_arg does not support Base.BottomRF in optimized IR aviatesk/JET.jl#863 with a reproducer.

This PR keeps the checks live on every Julia version instead of marking them broken:

• On Julia < 1.13: unchanged — the @report_opt bounds (0 reports for HCubatureJL, ≤2 for VEGAS) still run.
• On Julia ≥ 1.13 (including prereleases, via VERSION < v"1.13.0-"): the same solver calls are checked with @report_call — JET's call analyzer over unoptimized IR, which is unaffected by the optimized-IR issue and catches method errors / undefined-variable defects through the whole path — plus a native-inference assertion that solve infers to SciMLBase.IntegralSolution{Float64}.

The four opt-mode testsets that still pass on 1.13 (QuadGKJL, SampledIntegralProblem ×2, scalar infinite bounds) are left strict on all versions.

Alternatives considered

• broken = VERSION >= v"1.13": silences the only coverage on these paths for every 1.13.x release — a real method error or inference regression would go unnoticed until JET is fixed upstream. Rejected in favor of live checks.
• Hard version-gate with no replacement: same hole, plus no signal when upstream JET recovers.
• Rewriting substitute_u/substitute_v to dodge JET's widening: JET then reports deeper artifacts (it widens HCubature's integration point to Union{BitVector, Vector}, making the transformed-integrand closure Any internally). Not bounded or fixable package-side.

Test plan

• JET opt-mode solver paths testset on Julia 1.13.0 + JET 0.11.6: 8/8 pass on the new branch (@report_call clean, return-type assertions hold)
• Same testset on Julia 1.12.7 + JET 0.11.6: 6/6 pass on the JET opt-mode branch (unchanged behavior)
• Base.code_typed on solve(prob, HCubatureJL()) is identical on 1.12.7 and 1.13.0 — the Union{Vector{Float64}, Vector{Int64}, Vector{Real}} JET reports is absent from Base IR
• runic clean on test/qa/qa.jl
• CI QA group

🤖 Generated with Devin (model: SWE-2 Max)
Session: local transcript — /home/crackauc/.local/share/devin/cli/summaries/history_049e4e367a5d4d70.md (no shareable URL for this harness)

[ChrisRackauckas-Claude]

CI note: the Build and Deploy Documentation failure (missing_docs for the parametric CommonSolve.init docstring) is pre-existing on master — the last several master docs builds fail identically — and is fixed by #372, whose docs job is green. The typos failure was a flagged word in this PR's comments and has been fixed.