Revisit/memory qualitative: full-res per-method video panels for 6 memory benchmarks (mv/mb x dl3dv/spatialvid/kubric4d, ~10 clips each), lockstep SYNCHRONIZED playback controller on ALL qual panels, and per-clip return-frame comparison (generated-return | observed-GT + reused return/self PSNR/SSIM/LPIPS). Incremental: all prior sections (eval50/nvidia/spatialvid/kubric/lh qual, memory tables, difficulty hero, pose-free curves, fairness, coverage, glossary, story) preserved. [root]

Qualitative: FULL-RESOLUTION individual per-method clips in CSS grid (replace downscaled composited grids); native res, crf20, lazy-load. Non-qual sections preserved.

DEFINITIVE refresh: memory axis 9 methods/family (52 cells) — FrameCrafter folded into all 6 memory benchmarks (M1a/b/c revisit + M2a/b/c moving-back); hero 118 scored / 39 pending; verified-live upload [root mirror]

Memory axis: spmem now scored on all 6 memory benchmarks; SpatialVID M1c/M2c filled; cross-scene headline + degrade SVG (root mirror)

Incremental refresh: fold complete memory results (spmem 4 families) + new headline (root mirror)

Refresh root mirror + zip with Memory axis + LH-SOTA/ft_lh updates

Add Avg inference-time column (SLURM sacct) to quantitative tables + §5 runtime summary; infer_time.json

Remove DyCheck; add Kubric-4D angle-binned difficulty-curve hero (inline SVG)

Scene-extrapolation benchmark suite viz (live snapshot 2026-06-28): coverage matrix + quantitative tables + qualitative video panels