0.6.2rc2 release-candidate acceptance

This candidate includes the large-build, remote-registration and solver-development repairs collected since rc1. It preserves study deadlines, frozen reference identities and independent scientific acceptance. Operational recovery does not qualify a solver or establish a physical result.

Registration recovery on a live campaign

A read-only probe through the campaign’s frozen SSH transport took 66.22 seconds for 33 installed capabilities. Worker, code and configuration identities matched and the experiment launcher was available. Both blocked diagnostic instruments were present. The previous hardcoded 60-second registration deadline was insufficient.

The coordinator received a compatible registration-only patch. A graceful maintenance pause preserved the original deadline and numerical worker installation. The director then granted 120 seconds, one attempt, citing the measured latency. The worker registered the previously blocked instrument in 65.87 seconds, with all frozen identity checks passing. The new instrument still requires its applicable methods review before it can support scientific evidence.

Current limits, delegation, elapsed probe times, errors and director grants appear in the research monitor. Full attempt and adjustment records remain on disk. Every probe and retry respects the compute cutoff and protected report reserve. New studies enable director tuning; older studies require explicit delegation. See registration recovery.

Regression checks

  • 72 focused execution, registration, director, resource and browser checks passed. These include eight desktop/mobile monitor checks.

  • 41 compatible checks passed against the older coordinator source with the surgical registration patch. The live numerical worker was not upgraded.

  • 64 additional service, validation, finalization, waiting and study-launch checks passed.

  • The broad local run recorded 1,368 passes, five skips and two failures before the final status-reader repair. One failure exposed missing storage fields in a partial study manifest. It was fixed without inventing admission limits; the affected usage, registration and monitor checks passed in a 24-test rerun.

  • The other failure is the previously documented local FLASH/MPI sandbox smoke failure: its installed runtime aborts without producing HDF5 output. It also appears in rc1’s acceptance record. Its failure is retained; optional installed scientific runtimes are separate from the clean release suite.

  • All 25 public-server checks passed after adding the deployed version to /healthz.

  • All 14 DSH integration checks passed. Tracked Python source lint and JavaScript syntax checks passed. Documentation built with warnings treated as errors and its local link check reported zero failures.

The test sets overlap; their counts must not be summed into a single coverage number. Release automation performs the complete clean-source checks separately.

Installer portability and distributions

Issue #16 reports CRLF shell-script startup failures on WSL. The GitHub blob used LF, but the repository did not prevent checkout conversion. Shell scripts now enforce eol=lf. A regression test creates a core.autocrlf=true checkout and directly launches the ITER bootstrap; all five ITER integration checks passed. The issue response distinguishes checkout conversion from a genuinely missing Bash installation and provides an existing-checkout repair.

The source distribution, wheel and checksummed workspace bundle build successfully. A fresh-home installation required no agent credentials or system-package changes; the GUI, skills and HDF5 reader worked. Package, lock, citation, fallback and installer metadata consistently identify 0.6.2rc2. Private campaigns, credentials and operator device assets are excluded from the distributions.