[ChrisRackauckas]

🚧 UNREVIEWED — awaiting review by @ChrisRackauckas. Filed by an AI agent running as @ChrisRackauckas; Chris has not seen this.
Harness: Devin CLI 3000.11.92 (diagnosis and fix), Claude Code 2.1.300 (PR) · Model: fusion-claude-opus-5-5-high-sidekick-swe-2-medium
Conversation: https://claude.ai/code/session_01VZjwf7Ux9YtTzwHGHMqnsq (fleet master)

This fixes a failure on the default branch's CI. The diagnosis and the evidence below come from the investigating agent's report; "confirmed by running" and "read in logs" are kept apart there.

Failure 1: QA (Julia 1), Aqua "Persistent tasks" fails with Unable to locate Project.toml in .../RandomizedLinAlg/t4i9M

Error (read in logs)

Persistent tasks: Error During Test at ~/.julia/packages/Aqua/yUvBM/src/persistent_tasks.jl:40
  Test threw exception
  Expression: !(has_persistent_tasks(package; kwargs...))
  Stacktrace: ... Aqua.manifest_entries (persistent_tasks.jl:88/116/119) -> precompile_wrapper (142) -> has_persistent_tasks (51)
Some tests did not pass: 20 passed, 0 failed, 1 errored, 0 broken.

Julia 1.13.2, Aqua v0.8.18, SciMLTesting v2.13.2, RandomizedLinAlg v0.1.0. All Core jobs (lts, 1, pre, x86) pass. The QA group fails only on this check.

• The reported PR run: https://github.com/SciML/ModelOrderReduction.jl/actions/runs/38812682337/job/116689874438 (PR branch jet-0.12)
• The same error on the main push run for df48976: https://github.com/SciML/ModelOrderReduction.jl/actions/runs/38819777462/job/116713607588

Cause: upstream (Aqua 0.8.17+ combined with a REQUIRE-only dependency)

• Aqua v0.8.17 (2026-09-17, JuliaTesting/Aqua.jl PR #394 "Make test_persistent_tasks respect the environment") added manifest_entries. It builds the wrapper manifest by reading every dependency's Project.toml, and it calls error when one is missing. v0.8.18 keeps that behaviour. I read this in Aqua/yUvBM/src/persistent_tasks.jl.
• RandomizedLinAlg's only registered release, v0.1.0, contains REQUIRE and no Project.toml. I confirmed this by listing ~/.julia/packages/RandomizedLinAlg/t4i9M. Upstream master has a Project.toml but was never tagged. RandomizedLinAlg issue Proper Orthogonal Decomposition (POD) method #18, which asks for a release, has been open since 2026-10-05 with no response, and the repo has not been pushed since 2023.
• Timing: the last green main Tests run was on 2026-09-16 (https://github.com/SciML/ModelOrderReduction.jl/actions/runs/35146247645), one day before Aqua 0.8.17 was released. There were no main Tests runs between that one and df48976.
• Upstream tracking: Aqua issue Persistent tasks fails in some other packages JuliaTesting/Aqua.jl#404 is open. A collaborator says he is trying a "skip with a warning" approach, but there is no open PR for it in Aqua.
• Confirmed by running: on unmodified main (df48976), GROUP=QA julia +release --project -e 'using Pkg; Pkg.test()' reproduces the same error with Aqua v0.8.18 (20 passed, 1 errored).

Existing PRs

None. The open PRs in SciML/ModelOrderReduction.jl are #149, #88 and #27, and none of them touches this. There is no fix PR in JuliaTesting/Aqua.jl or JuliaLinearAlgebra/RandomizedLinAlg.jl.

Fix

The package used only RandomizedLinAlg.rsvd. The branch replaces it with a local implementation of the same algorithm (Halko, Martinsson & Tropp 2011, Algorithms 4.1 and 5.1: Gaussian range finder, QR, then the SVD of Q'A). This removes the REQUIRE-only dependency. No public API changes and no new dependency: randn is exported by Base, and qr!/svd! come from LinearAlgebra, which is already a dependency. RandomizedLinAlg is MIT.

• Project.toml: RandomizedLinAlg removed from [deps]/[compat].
• src/DataReduction/POD.jl: local _rsvd (throws ArgumentError when n + p exceeds the row count, as before). The reduce!(::RSVD) docstring Throws section is updated.
• src/ModelOrderReduction.jl: imports qr! and svd! from LinearAlgebra.
• src/Types.jl: the RSVD docstring no longer refers to RandomizedLinAlg.
• test/DataReduction.jl and test/Project.toml (test-only Random): new testset "POD - RSVD". For an exactly rank-3 40×25 matrix it checks that the RSVD spectrum matches svdvals (rtol 1e-10), that rbasis*rbasis'*A ≈ A, and that an oversized n + p throws ArgumentError. This tests the numerics of the replacement. It is not a test that fails before the fix. The test that fails before and passes after is the existing Aqua persistent-tasks check:

Before (main df48976, log scratch/qa_before.log):

Persistent tasks: Error During Test at .../Aqua/yUvBM/src/persistent_tasks.jl:40
  Unable to locate Project.toml in ~/.julia/packages/RandomizedLinAlg/t4i9M
ERROR: LoadError: Some tests did not pass: 20 passed, 0 failed, 1 errored, 0 broken.

After (branch, Aqua v0.8.18, log scratch/qa_after.log):

QA/jet_tests.jl |    5      5
QA/qa.jl        |   21     21
     Testing ModelOrderReduction tests passed

Other checks I ran on the branch (Julia 1.13.2): GROUP=Core passed, all 8 files, DataReduction 18/18. docs/make.jl built with doctests and only pre-existing warnings. Runic --check and typos are clean on the changed files. Not run locally: QA/Core on Julia lts (1.10) and the Downgrade workflow. I did check that bare randn resolves on 1.10.12.

An alternative I did not take: cap Aqua = "0.8, <0.8.17" in test/qa/Project.toml. That pins QA tooling to an old release and leaves the unmaintained, REQUIRE-only dependency in place.

Links

• Failing PR run job: https://github.com/SciML/ModelOrderReduction.jl/actions/runs/38812682337/job/116689874438
• Failing main run job: https://github.com/SciML/ModelOrderReduction.jl/actions/runs/38819777462/job/116713607588
• Last green main Tests run: https://github.com/SciML/ModelOrderReduction.jl/actions/runs/35146247645
• Aqua upstream issue: Persistent tasks fails in some other packages JuliaTesting/Aqua.jl#404
• Aqua PR that introduced manifest_entries: Make test_persistent_tasks respect the environment JuliaTesting/Aqua.jl#394
• RandomizedLinAlg release request: Tag a release that includes Project.toml JuliaLinearAlgebra/RandomizedLinAlg.jl#18

---

🤖 Posted by an AI agent — harness: Devin CLI 3000.11.92 (fix), Claude Code 2.1.300 (PR) · model: fusion-claude-opus-5-5-high-sidekick-swe-2-medium / claude-opus-5-5
Conversation: https://claude.ai/code/session_01VZjwf7Ux9YtTzwHGHMqnsq (fleet master)

🤖 Generated with Claude Code

Independent review (added by the fleet master)

Risk assessment

• Risk: low
• Blast radius: RSVD backend implementation and removal of a direct dependency; public POD/RSVD constructors and reduction behavior remain compatible.
• Evidence: Compared against RandomizedLinAlg v0.1.0 source. Independently ran 63 assertions on each of Julia 1.10.12 and 1.13.2: identical-seed U/S/V parity and element types for Float64, Float32, complex, tall/wide matrices; rank-3 reconstruction/spectrum and oversampling bounds. Unchanged local GROUP=QA on Julia 1.13.2 passes: JET 5/5, QA 21/21, including persistent tasks. PR QA/Core CI remains pending. Read the default-branch failing QA log independently: Aqua 0.8.18 errors on the missing RandomizedLinAlg Project.toml.
• Independent review: codex gpt-6.1-sol rated it low, high confidence in algorithm equivalence: the local helper preserves the released Gaussian range finder and projected SVD.
• Merge: auto-merge candidate — conditional on green QA, Core, downgrade and docs; no numerical or public-API defect found.

---

🤖 Section added by an AI agent — harness: Claude Code 2.1.300 · model: claude-opus-5-5 (review by codex-cli 0.162.1, gpt-6.1-sol)
Conversation: https://claude.ai/code/session_01VZjwf7Ux9YtTzwHGHMqnsq