0.6.2 release acceptance

This stable release includes the 0.6.2 candidates’ project/workspace collections, director recovery and storage controls, explicit experimental solver development, and subsequent study-extension, tool-selection and research-feedback repairs. It also adds the experimental Windows application runner from PR #21.

Live CST commissioning

The contributor reported live validation of commit fd86f5d972345aefad0ed0a96cc821ea34b58ff4 on CST 2024.4 with Windows Python 3.9.15. This is contributor-reported commissioning on that environment:

  • Interface checks and saved Two Coils results could be read.

  • Fresh magnetostatic solves at r=2 and r=1.5 reproduced the previously verified inductance/energy baselines, with maximum relative differences of 2.03e-15 and 5.25e-15. These measure baseline repeatability, not independent physical accuracy.

  • Queue serialization, lifecycle commands and hash-verified artifacts passed.

  • A native Python script opened the copied project, returned its result and saved/closed successfully. Cancelling another active script returned a confirmed cancelled receipt and wait exit code 130.

  • All 44 original project/model files remained unchanged. Test-owned processes exited while an unrelated existing CST session remained open.

The contributor disabled automatic CST update notifications locally to obtain unattended startup. The bridge does not automatically dismiss vendor dialogs. SOLIDWORKS model commissioning remains outstanding; COMSOL execution is not implemented. Windows jobs remain trusted operator-side exploratory processes.

Release preparation checks

  • 97 focused shared-runner, CST and installer tests passed.

  • Native Windows Python 3.9.13 and 3.12.10 passed 13 shared-runner checks each. The legacy entrypoint passed eight checks per interpreter, with five common-interface-only checks skipped.

  • All 14 DSH integration checks passed after the targeted dependency update. The pinned DSH version is retained; the lockfile now selects MCP SDK 1.32.1, sharp 0.35.5 and the matching sharp platform dependencies. The npm audit reported zero findings on 16 October 2026.

  • Source-to-wheel builds include the common Windows modules and discoverable application skill. The package does not redistribute vendor applications.

  • A fresh-home installation downloaded the workspace archive and checksum file, opened the GUI and loaded skills and HDF5 support without agent credentials or host system-package changes.

The installer verifier previously discovered the source checkout above dist/install.sh and bypassed the archive download. It now stages its bootstrap outside the checkout and requires both asset requests. The existing corrupted checksum regression continues to require failure before installation.

A post-merge Windows 3.9 CI fixture remained queued; its child logs were not retained, so that run’s precise cause is unconfirmed. A separate native reproduction demonstrated that a Windows reader denying delete sharing makes receipt replacement fail with WinError 5. The release adds bounded retries for Windows access/sharing errors, preserves the prior complete receipt, and tests the actual held-file conflict. Failed native fixture assertions now include supervisor and worker logs. This changes receipt handling, not the CST solver recipe.

Automated publication gate

The CI workflow runs the complete Python suite, documentation with warnings treated as errors, DSH integration, native Windows checks, recorded demos, deterministic campaigns, schema verification, packaging/install checks and the fixed CPU WarpX qualification matrix. The release workflow repeats the release suite, enforces the tag/version match and verifies the downloaded installer bundle before attaching assets and publishing to PyPI.

Historical acceptance and scientific campaign records keep their original software identities and outcomes. A stable package release does not qualify an experimental solver or change unresolved scientific findings.