idle

## Last work — 2026-05-14 18:03 PDT — wrapfig back to p3
Cameron clarified: wrapfig should be on p3 (after the per-pixel method figure), not p2.

Final layout:
- p1: Fig 1 overview
- p2: Fig 2 real_results (top, full-width)
- p3: Fig 3 method_perpixel (top, full-width) + Fig 4 method_height (wrapfig right-side, 0.33\linewidth, wraps §3.2 body text)
- p4: rest of §3 + §4 Experiments
- p5/6/7: Fig 5/6/7 ood/video/pretrain

Wrapfig source anchored right after `\subsection{Pixel-Aligned Heatmap Volume}` header, before §3.2 body. Numbering now restored to figure_maker's original spec (Fig 3 = perpixel, Fig 4 = height).

## Pipeline notes — /paper hang investigation
- omidlab.net/paper serving fine post-reboot tunnel restoration (200 OK throughout)
- "Hang" Cameron reported was an internet error on his end → stale browser/CDN cache
- One transient compile issue: after moving the wrapfig source, watcher's `tectonic` single-pass left `Figure ??` unresolved cross-refs in the PDF. Manual `tectonic main.tex` did the extra pass needed and updated the aux file. Subsequent watcher compiles produce 0 unresolved refs (verified via `grep -c "Figure ??"`).
- Possible follow-up: if structural edits (moving wrapfigs across pages) recur, the watcher's single tectonic invocation may again leave `??` until next compile. Tectonic does internal multi-pass but seems to bail under some aux-file regenerations.

PDF served at omidlab.net/paper, Last-Modified 2026-05-15 01:03:05 GMT.