Fit Page Limit

Two sides of the same problem: a paper that is over the limit and needs careful cuts, or a paper that is under and needs useful substance rather than filler. This skill maps the draft section by section, then gives the author a prioritized plan for whichever direction they need.

It coordinates with polish-prose for sentence-level tightening, refactor-structure for larger reorganization, polish-tables-figures for visual layout, and preflight-check when the page limit comes from a venue rule. Run it after the draft is roughly complete.

When to use

• The draft is over the limit and the author needs to know where the bulk is.
• The draft is under the target length and the author wants to use space well.
• The author asks for a length map: "where is all my length going?"

Inputs

• The .tex source.
• The real compiled page count, if available, to calibrate the estimate.
• The target: a venue page limit or a length the author names.
• The venue CFP or profile, if the target comes from a venue.

Process

1. Map the length. Run the analyzer from this skill directory; do not eyeball where the pages went.

   python3 scripts/section_budget.py paper.tex --target-pages 10 --current-pages 11
   

   It reports per-section words, percentage of the paper, estimated pages, and the over/under gap. --current-pages calibrates words per page to the author's actual compile. Use --json for machine-readable output.

2. Confirm what counts. Re-verify the venue's limit and exclusions against the live CFP before relying on appendix moves or excluded sections. A stale page limit can desk-reject the paper.

3. Pick the right branch. Use references/resize-tactics.md to turn the map into a value-aware resize plan.

If over: compress value-aware, never blind

Rank sections by compressibility x low impact, and cut there first:

• Redundancy: the same point repeated in the intro, method, and conclusion.
• Over-long related work: condense paragraph-per-paper prose into clusters that state only the delta that matters.
• Verbose method/setup: tighten with polish-prose; cut walkthroughs a figure already shows.
• Move, do not delete: push proofs, extra results, or implementation detail to an appendix only when the venue excludes it.
• Figures/tables: resize, merge subfigures, or drop a redundant baseline column with polish-tables-figures.

Protect contributions, headline results, and every citation. Present estimated savings per cut and let the author approve before editing. Never delete a result, claim, or reference to hit a number.

If under: expand with substance

A short, complete paper can be fine; say so. If there is real room to strengthen it, suggest substance: a missing ablation, deeper error analysis, limitations or threats to validity, clearer motivation, a worked example, or a related-work gap. Flag suggestions that would merely pad. Hand drafting to the writing skills.

Output

A length map and prioritized resize plan:

• Over: ranked cut list, estimated savings, and protected material.
• Under: substantive additions ranked by value.

Write durable outputs to the user's local paper-workspace/writing/ directory when a workspace exists. The skill recommends and explains; the author decides and edits.

Guardrails

• Never delete results, claims, or citations to fit.
• Never pad to fill.
• Page numbers are estimates; tell the author to compile and check the real count.
• Re-verify venue limits and exclusions against the live CFP before advising appendix moves.
• Copilot, not pilot: propose the plan; the author approves and edits.

Memory

Uses the shared .paper-memory/ convention documented by paper-profile.

• At start: read lessons.md for recurring overspend patterns, such as "related work always runs long".
• At end: append `date - fit-page-limit - over/under - where the length went
  • what was cut/added` after deduping.
• Create .paper-memory/ on demand in the user's paper directory. It is local only and should usually be ignored by git.