Cluster scheduling, parameter studies, and the workspace asset store are now supported, as are all three study types (grid_independence, timestep_independence, sensitivity) and all four input import modes (copy, hardlink, reference, reflink). On a Slurm cluster, picurv cancel stopped a running solver, cancel --graceful made it write an off-cadence checkpoint that run --continue resumed from, post jobs chose their steps when they started, array studies of every type honoured max_concurrent_array_tasks and aggregated a row per member, and each import mode behaved as documented on Lustre. A successful reflink has not been observed; the filesystems tested refuse it.
sweep --continue works for studies inside a workspace. It looked for the study's base configurations in the workspace rather than in the study, so it failed for every such study.
submit --stage post-process no longer waits on a solve job that has already finished. Slurm forgets a finished job and then rejects any dependency on it; a completed solve is now not waited for, and a failed one is refused unless --force is given.
simulator --version and postprocessor --version now also name the PETSc they were built against (version, debug or optimized, arch, and directory). picurv version shows it, even for an executable that cannot start, together with the libpetsc the current shell would load, and marks a mismatch such as a debug build run against an optimized PETSc (undefined symbol: petscstack). Each run's software lock records it, and a job now stops if its executable was rebuilt against a different PETSc after staging.
Study metric tables and plots are documented at their real location, studies/<study_id>/output/analysis/. New troubleshooting entries cover an sbatch that rejects a job while exiting 0, and the petscstack error.