Fix AdaptiveSDE benchmarks

[ChrisRackauckas]

https://github.com/SciML/DiffEqBenchmarks.jl/blob/5acdeba99e5f43e2748c0beadf90a48233f2f373/markdown/AdaptiveSDE/AdaptiveEfficiencyTests.md

https://github.com/SciML/DiffEqBenchmarks.jl/blob/5acdeba99e5f43e2748c0beadf90a48233f2f373/markdown/AdaptiveSDE/qmaxDetermination.md

[ChrisRackauckas]

See #451 . The code is now updated and correct, but we need JuliaLang/julia#43270 in order for the benchmark machine to work correctly. Hopefully it gets into v1.8.x because a lot of other things need this to work well too with distributed.

[ChrisRackauckas-Claude]

AdaptiveSDE Benchmarks Status (January 2026)

Both AdaptiveSDE benchmarks remain broken as of January 2026:

AdaptiveEfficiencyTests

Error: Failed to precompile StochasticDiffEq package, leading to cascade of undefined variable errors (probs, gr, etc.)

qmaxDetermination

Error: Same precompilation failure, with additional UndefVarError for Random and probs

Root Cause

The StochasticDiffEq package fails to precompile in the benchmark environment, preventing the entire benchmark suite from executing. This appears to be a dependency or compatibility issue.

Impact

Neither benchmark produces any results or visualizations. The published pages show only error messages instead of the intended adaptive noise algorithm comparisons.

cc @ChrisRackauckas

[singhharsh1708]

@ChrisRackauckas i would like contribute on this also .

[ChrisRackauckas]

This one is a bit difficult. It requires probably changing the testing setup in general because I think the issue is that Weave.jl itself isn't compatible with distributed.

[jitendravjh]

Hi @ChrisRackauckas — I'd like to take this on, and went through both benchmark files on master plus the current SciML APIs to scope it.

AdaptiveEfficiencyTests.jmd — calls DiffEqBase.calculate_monte_errors, which appears to have moved out of DiffEqBase (now in DiffEqDevTools) and isn't in the benchmark's Project.toml. I'd port it to the current ensemble-error API and update the downstream .elapsedTime/.error_means[:final]/.error_medians[:final] accesses to match.

qmaxDetermination.jmd — relies on DiffEqMonteCarlo (removed/superseded by the EnsembleProblem interface, and also not in Project.toml); I'd port it over, and update the oval2 accessor from SDEProblemLibrary to its current name.

General — bump Project.toml/Manifest.toml, get both files weaving cleanly in CI, and regenerate the efficiency plots. This looks like the same shape as the previously-funded "Fix and Update the Simple Handwritten PDEs" benchmark set.

Two questions before I open a claim PR on small_grants.md:

1. Are these benchmarks still wanted as-is, or have they been superseded?
2. Would this qualify for the SciML Small Grants program, and if so what amount/reviewer should I list?

Happy to start as soon as it's confirmed — thanks!