SSB reports#
Dated records retain their original workload, backend, source revision and
acceptance limits. They are evidence, not a statement of current release speed.
A record that describes since-retired behaviour (the phase-variance search,
Nelder–Mead or finite-difference refinement, the mean_phase and
complex_wave estimators) says so in a Historical line at its top.
Native Metal SSB redraw precision at the lowest scan frequencies
Redraw at slider speed: q, -q symmetry for SSB and a drag window for regularized methods
Native Swift/Metal SSB redraw while dragging aberration controls
Lower CUDA SSB staging memory without changing numerical results
Sign of the odd aberrations in SSB reconstruction (2026-10-21)
Coma, trefoil and spherical aberration in the model-based SSB fit (2026-10-21)
The MPS standard SSB fit uses the agreement objective (2026-10-20)
Tilt-corrected SSB on CUDA: fit and preview speed (2026-10-20)
CUDA standard fit on the agreement objective, and one phase-variance search (2026-10-20)
Tilt-corrected SSB on Metal: preview and fit speed (2026-10-20)
MPS performance: what got faster, why the rest is slow, how to profile (2026-10-20)
MPS memory: where SSB’s gigabytes go and what fits which Mac (2026-10-20)
Reuse MPS loading buffers and decode detector regions directly
CUDA fine-grid kernel experiments, 23-24 October 2026#
These records measured in-kernel q, -q pairing, compact support and closed-sideband skips on a separate branch. Main took a different design (half-plane spectrum with the -conj mirror, bounded BF groups and drag windows), so the kernels they describe are not in the current source. The numbers and rejected ideas remain as evidence.