Packing and density

For each of the 36 variants (preset x length x template) the packer in scripts/build-resumes.ts fills the page rather than just avoiding overflow.

The algorithm

1. Rank. Every entry is a unit, scored by weight x recency. Units are ordered on-preset first (tagged for this preset), then off-preset, each group by score.
2. Admit. Walk the units in order, adding each with its two best bullets and keeping it only if the page still fits.
3. Grow. Walk the admitted units again, adding bullets one at a time (cap MAX_BULLETS) until the page is full.
4. Borrow. When on-preset content runs dry, off-preset bullets fill the rest so a narrow preset still reaches a full page.
5. Density. With content fixed, open spacing to the largest DENSITY step that still fits, filling the target exactly.
6. Collapse. A two-pager becomes a full one-pager only if two pages cannot be filled even at maximum density.

The result is ~95% fill on every variant, with two-pagers always two full pages.

Density band

DENSITY in scripts/build-resumes.ts sets the spacing multipliers. It scales line spacing and section, entry, and bullet gaps in the templates via the density value in the data. The band is deliberately narrow because resume typography research puts readable, ATS-safe line spacing at roughly 1.0 to 1.15; most of the range lives in the gaps, and margins stay fixed.

designed: [1.0, 1.08, 1.16, 1.25, 1.34]
ats:      [1.0, 1.05, 1.1, 1.16]

Bullets: tags and priority

• tags on a bullet decide on-preset relevance (which preset the bullet belongs to). A bullet tagged [all] shows on every preset.
• priority (lower is more important) orders bullets within an entry.
• On the full length, only on-preset bullets are shown, so per-preset variants of the same point do not stack.

Testing

npm run build-resumes

Check lib/resume-manifest.json for each variant's pages, and open the PDFs in public/resumes/. Both are gitignored build artifacts, regenerated every build.