Skip to content

[KAMEL] Ship and validate the production periodic integral-transport workflow #295

Description

@krystophny

What to build

Turn the completed periodic KIM and QL-Balance slices into a reproducible KAMEL workflow. Expose validated configuration through the Fortran and Python entry points, record complete HDF5 provenance, add end-to-end golden cases, and produce the scientific convergence and performance report required for human acceptance.

This implementation work is AFK-ready. Declaring the new physics workflow production-accepted remains an explicit human scientific sign-off after the report is available.

Acceptance criteria

  • Configuration explicitly selects periodic KIM, harmonic bounds, drift-kinetic electrons, integral ions, B-parallel source, compact-transition parameters, target-current guards, and benchmark mode.
  • Unsupported model combinations fail validation instead of falling through hidden hard-coded branches.
  • Python orchestration can configure and execute the complete coupled workflow.
  • HDF5 output records per-mode local windows, transition metadata, Br, B-parallel, electric fields, species current, unscaled/scaled tensors, shielding history, model names, Fourier convention, and algebra-generator hash.
  • One small production golden case covers periodic KIM, both species policies, B-parallel, target-current normalization, profile evolution, and restart.
  • A separate golden case covers the old/new ion drift limit and records all four residuals.
  • Automated scans cover Fourier/radial resolution, periodic-core and transition width, cyclotron harmonics, finite-Larmor-radius convergence, B-parallel channel isolation, and time-step refinement.
  • A representative central-layer comparison against global KIM is reported without treating global agreement as the sole correctness oracle.
  • Periodic coupling wall time and memory are measured against global KIM for representative target workloads and demonstrate the intended speed advantage.
  • Existing KiLCA, global-KIM, and default QL-Balance regression suites remain green.
  • User documentation explains the model boundary, validity region, normalization, benchmark, diagnostics, and restart procedure.
  • A review artifact summarizes tolerances, residuals, taper sensitivity, convergence, performance, and remaining scientific uncertainty for human sign-off.

Blocked by

Metadata

Metadata

Assignees

No one assigned

    Labels

    P2Normal: scheduled workReview effort 5/5enhancementNew feature or requestready-for-agentFully specified and ready for an autonomous agent

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions