leonarski_fandClaude Opus 5.5
1eda55ab38
The rotation-scale walk ran its probe passes on the run itself and put back a list of members afterwards (experiment, geometry, pre-pass mosaicity, scale). The list missed what an integrating refit probe leaves - bkg_starved_fraction_, last_rotation_result_ - and the spot-finding settings a first-pass ladder rung adopts were carried on purpose, with a check that threw the refined passes away when the walk had changed them. Each probe now runs on a copy of the run as the walk found it, and only its first-pass memos (an exact cache) come back; the run ends the walk as it began it, the restore lines and the spot-finding check go, and the refined passes are overturned only where a scale is adopted. The geometry walk's index_at_geometry probe runs on a copy the same way, so a probe the speculative one already answered (memo hit) and one run there (miss) leave the run identical. Measured before the change with a temporary state dump at the start of the refined passes, walk run against a run that skips the walk and applies the walk's k directly (8r5r, 9gdj, 9qw8, 2ygz adopt k != 1; 9t6s, lalanine walk first and keep k = 1): p.hkl identical on all six; the run state differed only in last_rotation_result_, bkg_starved_fraction_ and the memo count - leaks the canonical pass overwrote before reading, so latent. After (GPU, gpulock, shared machine): p.hkl and MTZ reflection data identical to the base binary on 26 ci-tier sets with a merge (+ both no-crystal controls, same verdict), on the 11 open sets whose walk adopts k != 1 or runs first, on 7 sets whose geometry walk takes a sub-step round, and on the one private set that adopts a scale. Catch2 [RotationScale] and Rugnux_FirstPassMemoMatchesRecomputation (verify_first_pass_memo through the walk) pass. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EizBKhTqrAYago9KJG3dA3
Jungfraujoch
Application to receive and analyse data from the PSI JUNGFRAU and EIGER detectors, together with
rugnux, its program for processing stored diffraction data - rotation and serial, macromolecular
and small-molecule - from images to merged reflections.
All documentation is now placed in docs/ subdirectory and for the current version hosted on Jungfraujoch Read The Docs page.