[janosg]

Problem

1. 22 Cartis-Roberts problems (arglale, arglble, bratu_2d, bratu_2d_t, bratu_3d, drcavty1–3, eigena, eigenb, the six flosp2*, hydcar6, hydcar20, qr3d, qr3dbd, yatpsq_1, yatpsq_2) and the More-Wild kowalik_osborne problem define their criterion with functools.partial but without mark.least_squares. (kowalik_osborne used mark.least_squares(partial)(...), which marks partial itself.) get_benchmark_problems therefore treats them as scalar problems, minimize raises InvalidFunctionError ("Scalar objective values need to be scalars"), and run_benchmark with error_handling="continue" silently counts them as failures. Four of them are in the estimagic set.
2. With scaling=True, noise_free_fun is the unscaled criterion while inputs["params"] are scaled. run_benchmark therefore computes wrong start values for failed runs and wrong criterion histories for noisy problems (e.g. 9972 instead of 24.2 at the start of rosenbrock_good_start).

Changes

• Wrap the partials in mark.least_squares.
• Build noise_free_fun like the criterion, with the same scaling but without noise.
• Tests: every More-Wild and Cartis-Roberts criterion is marked as least squares; noise_free_fun reproduces the start criterion at the scaled params; minimizing a partialled problem works; run_benchmark with noise and scaling starts at the correct value.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
...c/optimagic/benchmarking/get_benchmark_problems.py	93.95% <100.00%> (+0.04%)	⬆️
src/optimagic/benchmarking/more_wild.py	100.00% <ø> (ø)
...ests/optimagic/benchmarking/test_cartis_roberts.py	100.00% <100.00%> (ø)
...imagic/benchmarking/test_get_benchmark_problems.py	100.00% <100.00%> (ø)
tests/optimagic/benchmarking/test_more_wild.py	100.00% <100.00%> (ø)
tests/optimagic/benchmarking/test_run_benchmark.py	100.00% <100.00%> (ø)

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.