Paper: full review pass — figure correctness and typesetting

[isPANN]

Goal

Do a head-to-tail review of docs/paper/reductions.typ and the rendered PDF. Some figures are mathematically or visually wrong, and overall typesetting needs at least one cleanup pass before the paper is share-ready.

Scope

1. Figure correctness

   • Walk through every #reduction-rule and #problem-def example figure, check the picture actually matches the stated instance / witness / gadget construction.
   • Flag mislabeled vertices, wrong edges, stale drawings left over from earlier instance versions, inconsistent node colors vs. selected sets, etc.
   • Fix or re-draw; keep CeTZ conventions from .claude/rules/typst-drawing.md (name objects, connect by anchor, layering via on-layer).

2. Typesetting pass

   • Page breaks around figures / theorems — avoid orphaned captions, theorems split across pages when avoidable.
   • Consistent figure sizing and caption style across sections.
   • Math spacing ($...$ vs $ ... $), consistent use of arrow.r / arrow.long, display-name dict consistency.
   • Heading hierarchy, TOC cleanliness.
   • Verify citations render correctly (no @author? misses) and bib file is tidy.
   • Completeness warnings at end of file should be empty (or explicitly acknowledged).

3. Cross-checks

   • display-name dict vs. problem-def labels — no stale entries, no missing ones.
   • Every rule that appears in reduction_graph.json has a reduction-rule entry (the file already has a completeness check — make sure it's clean).
   • Overhead formulas in the paper match the #[reduction(overhead = {...})] in code.

Definition of done

• Every figure in the rendered PDF has been eyeballed against its stated instance/construction; issues either fixed or filed as follow-ups.
• No broken citations, no Typst warnings, completeness checks pass.
• A short change log in the PR description listing categories of fixes (figure fixes, spacing, citations, etc.).
• Reviewer sign-off from at least one other maintainer.

Non-goals

• Not a content/math-rewriting pass — this is correctness + polish, not restructuring sections.
• Does not include performance work (see Typst paper compile: finish optimization pass and verify output #1037).
• Does not include fixing underlying reduction bugs surfaced during review — those should be filed as separate [Rule] / bug issues (e.g. KSatisfiability → TimetableDesign reduction: gadget blowup 64·L² seems excessive #1036).

Related

• Typst paper compile: finish optimization pass and verify output #1037 (compile-time optimization; merge first to keep iteration fast)
• KSatisfiability → TimetableDesign reduction: gadget blowup 64·L² seems excessive #1036 (one concrete reduction with an obviously-wrong blowup)