SEIHouse Font Lab

Home of SEIHouse Sans (formerly SEIReader), the reading font of SEIHouse Productions LLC. It is built for reading hundreds of chapters on a phone: calm, soft, and distinctly SEIHouse.

The SEIHouse Display Engine is a version 0.1 workshop for album covers, posters, chapter titles, and other large text. It shares the Reader drawing engine and builder core; its four cut settings, live Lab and prebuilt fonts are under display/.

Font Lab homepage & Vercel

Open index.html for the workshop homepage: an editable specimen using the actual font files, Reader/Display workspaces, comparisons, chapter reading tests, and downloads. Both Labs can save browser drafts and download JSON settings on a normal website.

Import this repository into Vercel with Other as the framework and the repo root as the Root Directory. vercel.json builds the website with node build_site.mjs and serves only dist/site. There are no website dependencies or environment variables to configure. See website setup and draft usage.

Local preview: node build_site.mjs, then python -m http.server 8000 --directory dist/site and open http://localhost:8000. To design Display cuts locally, run python -X utf8 display/lab_server.py instead and open http://localhost:8000/display/lab/: it adds Build font, shipping a cut and motion approval to the Lab (how it works).

Everything in the font is drawn from rules in code. There are no hand-drawn files to maintain, so any change (weight, letter height, roundness, spacing) rebuilds the whole family consistently.

Installing in an app

Use the App distribution Actions run's seireader-app-packages download, or build the packages locally with python build_distribution.py and npm pack --ignore-scripts --pack-destination dist. No font-building dependencies are needed to package the checked-in fonts.

• Static assets: extract SEIReader-0.39-web-full.zip into your app's public assets and load its sans.css for font-family: "SEIHouse Sans".
• Vite/npm: install seihouse-seireader-0.39.0.tgz and import @seihouse/seireader/sans.css. This private package has zero dependencies and no installation scripts; it is not published to the npm registry.

The default uses ten full WOFF2 files: 712,464 font bytes (695.8 KiB) for all five weights and their real italics. That is 71.6% less than storing all seventy overlapping subsets. Regular plus italic uses 141,632 bytes; the browser requests the styles used by rendered text. The package includes no old versions, Lab, OTFs, reference fonts, proof images, or font engineering tools.

See APP-INSTALL.md for both installation routes, explicit weight-only bundles, how Development's NovelExpanded Reader uses both SEIHouse font packages, and how to update them there after a font release. The developer sections below describe the full source repository.

Display font Lab

Open display/lab/index.html to explore Soft, Edge, Ink, and Wide. Each design has its own OTF and WOFF2 files under display/fonts/<cut>/. Reader and Display now use the root engine.js and font_builder.py. Display cuts inherit the current Reader Latin, Vietnamese, Cyrillic and Greek repertoire, combining mark/local forms and numeric features while retaining their cut settings. Display fonts remain excluded from the SEIReader app ZIP/npm package. See the shared engine and Reader preservation gate and clean-shape construction and proofs. Titles use per-cut capital-height kerning and optional cpsp capital spacing. The Lab reads each built font's final spacing; see the title spacing gate and before/after proofs. Each cut now chooses its own letter designs. The Lab has individual controls for a/g, R, K/k, M/W, y, G, Q and open/closed 4/6/9; built fonts include the other forms through ss01–ss08. See the alternate mapping, cut defaults and native title proofs. All four presets default to plain W/w; crossed W is optional in ss04. Display embeds the Reader's ecosystem license metadata and includes six web subsets per cut through display/fonts.css, with every Latin-basic file below 25 KB. The phone Lab gates overflow at 360–430px. See the pre-Step-5 fixes, current specimens and delivery checks. The production pipeline discovers every cut JSON, supports opt-in extra weights and Oblique, and generates a finished-font specimen collection. The App distribution Action rebuilds and gates every declared face before uploading seihouse-display-fonts, including OTF/WOFF2, subsets, CSS, specimens and licenses. Current cut defaults remain Regular-only. Its Save cut button uses browser-local drafts on a normal website and retains the original database integration in the Claude host. Use Download JSON or Copy to bring a cut back to the font builder. Drafts stay on this browser/site.

To rebuild with the Python dependencies described below, run from the repository root:

python -X utf8 display/make_display.py display/cuts/soft.json
python -X utf8 display/build_display_page.py

For the entire collection, run python -X utf8 display/build_all_cuts.py. Use --output-dir dist/display-production for a separate candidate bundle; per-cut production.weights and production.oblique settings opt into styles. See Display build options and gates.

Choose edge.json, ink.json, or wide.json for another design. The builder uses Chromium and otfautohint; run it in the environment where those dependencies are installed. The explicit UTF-8 mode supports the supplied Unicode sources on Windows. The Lab generator checks the cut settings and font hash against its spacing export. Unbuilt geometry edits receive a draft-spacing label until rebuilt. Run python -X utf8 verify_display_shapes.py and python -X utf8 verify_shared_engine.py to gate all four rebuilt cuts and Reader preservation, and python -X utf8 verify_display_spacing.py to gate Lab/font title widths at a difference below 0.5%.

What's inside

For application integration, @seihouse/living-titles is a separate private runtime package with a ready-to-install tarball, ESM/TypeScript interfaces, a browser player, an optional React component and four static Display faces. The NovelExpanded mock consumes that same runtime. Audio remains caller-owned; menu and Reader Chamber systems stay separate. The existing Reader font package and font build interfaces remain unchanged.

Folder / file	What it is
index.html, site/, build_site.mjs, vercel.json	Workshop homepage, browser draft/download helpers, and an explicit static website build for Vercel
display/	SEIHouse Display 0.1: four cut settings, live Lab, shared-core entry point, kerning maps and OTF/WOFF2 fonts
fonts/	The finished fonts: 5 weights (Light, Regular, Medium, SemiBold, Bold), each upright and italic, as full .woff2 and .otf files plus seven WOFF2 subset deliveries per style
fonts.css	Ready-to-use @font-face rules for all 70 Latin and script WOFF2 subsets
fonts-full.css	Recommended app stylesheet for the ten full web fonts; no Unicode face splitting
package.json, build_distribution.py, verify_distribution.py	Runtime allowlist, deterministic ZIP delivery, dependency-free npm package, and actual shipped-payload audits
docs/APP-INSTALL.md	App installation, measured sizes, Development's NovelExpanded Reader integration, and the steps for updating the fonts there
lab/index.html	The Lab: the one page for testing and tuning the font (Reader Chamber, weights, languages, symbols, spacing tools, emoji comments demo)
lab/comparison.html	The same reference preview: 0.39 beside preserved 0.34, Literata, and Rubik, with the expanded language selector, NFC/NFD alphabets, local forms, and the established reading samples
lab/reading-test.html	Three full chapters for sustained reading, with Day/Night and weight controls plus locally saved feedback
docs/HOW-TO-USE.txt	Copy-paste instructions for a coding agent to add SEIReader to an app
docs/HEALTH-CHECK.txt	Results of Google's FontBakery checks, version by version
docs/DEVICE-TEST.md	Pass/fail checklist for iPhone, Android, Windows, and Mac browsers
settings.json	The main settings: weights, letter height, roundness, spacing, reading setup
phase4.json, phase4_support.py	Current additive release metadata and web-face order, kept separate from design settings
engine.js	Shared Reader/Display rules for every character, mark, alternate and symbol
font_builder.py	Shared outline, spacing, OpenType layout and hinting core
make_fonts.py	Thin Reader entry point: builds all 10 styles as full OTF and WOFF2 files
verify_shared_engine.py, verify_display_shapes.py	Reader preservation, Display shaping and all-glyph 400px shape/delivery gates
build_subsets.py	Uses pyftsubset to split each style and writes fonts.css
verify_phase2.py	Shapes the optional figures/fractions and checks all current subset files
latin_layout.py, verify_latin.py	Build and audit Latin composition, base/mark/ligature attachment, stacking, dotless forms, and preserved 0.32 outlines/spacing
languages.json, language_coverage.py, verify_phase4.py, verify_phase4_engine.mjs	Pinned CLDR inventories, modern script scope, and additive coverage, composition, collision, preservation and web delivery audits
verify_lowercase.py	Checks the scope of the 0.29 curve pass against preserved 0.28: all advances, unchanged outlines, counters, heights, and settings
verify_spacing.py	Checks the 0.31 pass against preserved 0.30: every contour and metric, permitted pair changes, shaping, accent classes, joins, hints, subset spacing, and collisions
verify_texture.py	Checks 0.32 against preserved 0.31: permitted contours, stable advances and letter rhythm, figure counters, reading weights, hints, shaping, and collisions
build_page.py	Builds the Lab page with the current fonts inside it
build_comparison.py, comparison_template.html	Rebuild the self-contained reference preview with embedded fonts; no network access needed
docs/DESIGN-GOALS.md	The approved design direction, reference measurements, and reading experiments
docs/LATIN-FOUNDATION-0.33.md	The first multilingual expansion step: accent inventory, quotation conventions, delivery, validation, and coverage limits
docs/LANGUAGES-0.34.md	The first language batch: scope per locale, 94 additions, local behavior, usage, inspection, data provenance, and remaining work
references/	Official Literata/Rubik comparison fonts with OFL licenses and source hashes, plus the separate Unicode license for CLDR locale data
emoji_demo.py	Original SEIHouse emoji art used in the comments demo
kern_base.json, kern_auto.json, kern_styles.json, kern_languages.json	Pair spacing: hand-set pairs, automatic and finished per-style maps, and Hungarian-only capital adjustments used by the Lab
render_small_text.py, render_windows.ps1	Optional Windows WPF small-text proof renderer; uses HarfBuzz shaping and the actual glyph indices, with no font installation
old/	The 0.6 version for the Lab and all ten styles of 0.27 through 0.33 as preserved baselines; 0.31 through 0.33 also include their thirty subsets

Rebuilding

Needs Python 3 and:

pip install -r requirements.txt
python -m playwright install chromium

Then:

python make_fonts.py        # builds all 10 full styles, 70 subsets, and fonts.css
python build_page.py        # builds lab/index.html with the new fonts
python build_comparison.py  # builds lab/comparison.html with candidate, baseline, and references
python verify_phase2.py     # checks OpenType shaping and subset coverage
python verify_phase4.py     # checks all additions and exact preservation against the Phase 4 starting commit
node verify_phase4_engine.mjs # checks old SVG output across weights, italics and contrast
python verify_phase4.py --custom-spacing # optional real-font Lab override regression
python verify_latin.py old/0.33 # historical foundation audit against preserved 0.32
python verify_texture.py old/0.32    # historical 0.32 texture/weight scope against 0.31
python verify_lowercase.py old/0.31  # historical curve preservation against 0.28
python verify_spacing.py old/0.31    # historical 0.31 spacing scope against 0.30
python build_distribution.py       # runtime-only app ZIP; standard library only
npm pack --ignore-scripts --pack-destination dist # private, dependency-free npm tarball
python verify_distribution.py --npm dist/seihouse-seireader-0.39.0.tgz
python verify_license.py           # legal metadata in all current full/subset fonts

The historical scope audits target their recorded releases; the current additive gate is verify_phase4.py. The 0.35 stroke-join audit requires that release's original font set.

To apply changes saved from the Lab (copied with "Copy instead of Save"), put them in a file and run python make_fonts.py my_changes.json.

Optional native Windows proof: python render_small_text.py path/to/output 96 --languages. This produces Day/Night PNGs at 13/15/17px in Light, Regular, and Medium, including real italics. See the checked-in Day and Night proofs. Add --full-family to proof all five weights at 20px, upright and italic; see the family Day and family Night images. Use --latin for foundation marks, --texture for punctuation/figures, or omit the flags for broader word-spacing samples. The language proof uses 1.5 leading with at least 21px baseline separation. These are WPF rasterization evidence; physical browser/device results remain deferred.

For the 0.35 export repair, see stroke-join notes and the Chromium Day / Night proofs.

The final paragraph-spacing inspection records the sentence-ending, apostrophe and narrow-letter candidates in full paragraphs, with Day/Night proofs across all ten styles. It found no demonstrated spacing defect and preserves the 0.35 fonts. Open Paragraph spacing in each reference preview card to inspect the exact text with the existing reading controls.

What SEIHouse Sans has (version 0.39)

• 5 weights with real italics (handwriting-style italic letters, not a tilted copy)
• Thick and thin: horizontal strokes 12% thinner than vertical ones
• Tall lowercase letters exactly capital height
• Automatic pair spacing for capitals, punctuation, quotes, brackets, and gently corrected lowercase pairs, plus hand-set pairs
• Rounder upright letters and slightly wider forms; raised straight and curly quotation marks
• Individual reading curves for a, e, c, and s; modestly narrower a/e, more side space for i, and a 0.220 em word space. Heights, nominal weights, contrast, squared caps, and the handwriting-style italic construction carry forward from 0.27
• Related h/n/m shoulders, the u lower curve, and b/d/p/q bowls refined to sit consistently beside a/e/c/s. All advances and global settings carry forward from 0.28; real italics retain their lower branches and pen-like exits
• Joined fi and fl
• Focused optical spacing for the reported snag words and ri/rn/cl/li, measured separately for every weight and italic. Joined fi/fl receive spacing beside neighbors; all glyph advances carry forward, with extra separation at crowded tt/ry pairs
• Broader rhythm inspection for minimum, murmur, river, climate, parallel, and everywhere, plus supporting r/v/w/y and narrow-letter cases. Balance ll, protect yw/tw separation, and open crowded italic ur/um. All earlier focused pairs and all character advances are preserved. See the inspection notes
• Quieter punctuation and smoother figures, including updated tabular/small figures and fractions. Raised quotes stay raised; mark/figure pairs retain minimum ink separation, including accented letter classes. See the texture notes
• Firmer Light and slightly lighter Medium letter strokes for small reading sizes, with Regular/SemiBold/Bold letter drawings, symbols/icons, nominal controls, and all advances unchanged. These are static refinements at every size, with no optical-size axis
• Optional tabular figures (tnum), superscript and subscript figures (sups, subs), and stacked fractions (frac); Unicode superscripts/subscripts and ½ ¼ ¾ are also included
• Eighteen zero-width combining accents with automatic ccmp, mark, and mkmk layout. Existing accented text has identical NFC/NFD shaping; new combinations support above/below attachment, stacks, real italic placement, and dot removal under above accents on i/j
• Local low/reversed quotes ‚ „ ‛ ‟, lexical apostrophes ʻ ʼ, a dotted circle ◌, and U+2009 thin/U+202F narrow nonbreaking spaces. All 295 existing drawings, metrics, settings, and pair values are preserved from 0.32. See the Latin foundation notes
• Ninety-four additional encoded letters complete the pinned CLDR main/auxiliary/index inventories and case variants for Polish, Czech, Hungarian, Romanian, Turkish, Hausa, Kurdish (Kurmanji, Latin), Māori, Igbo, and Uzbek (Latin). New side carons, hooks, barred letters, Igbo tone stacks, Turkish dotted fi, Romanian comma forms, and Hungarian capital digraph spacing follow the existing design. All 323 old outlines/metrics and nonnumeric pair values match 0.33. Numeric periods/commas have explicit separation on both sides, including equal adjustments for tabular figures; digit drawings, advances, and digit-to-digit spacing stay unchanged. See the language notes
• Seven WOFF2 deliveries per style: Latin basic, legacy Latin extended, symbols/icons, latin-ext-2, cyrillic, greek, and vietnamese. The generated fonts.css registers all 70; preferred script faces include complete Latin text, punctuation and combining marks
• Screen tuning (alignment zones and Adobe autohinting) for crisp small text
• Established Latin coverage: English, Spanish, French, Portuguese, German, Italian, Dutch, Catalan, Danish, Norwegian, Swedish, Finnish, Icelandic. Additional audited inventories: the ten locales above, Slovak, Croatian, Slovenian, the six Cyrillic locales, monotonic Greek and Vietnamese in Phase 4. Alphabet coverage is separate from native-reader and physical device certification; other scripts/spelling traditions may need fallback
• Web-novel and system-screen symbols, music marks, player controls, icons, and the SEIHouse Ⓢ brand mark, with private in-app codes (U+E000–E00F) so phones can't swap in color emoji
• FontBakery: 0 failures and 0 warnings in OpenType and offline universal checks on all 10 full OTF styles; online-only universal checks remain unverified in this session

Phase 4: additive language expansion

Step 1 (0.37) completes all 128 Latin Extended-A characters. Polish, Czech, Slovak, Hungarian, Romanian, Croatian, Slovenian, and Turkish are covered, including İ ı Ğ ğ Ş ş Đ đ Ĺ ĺ and the established apostrophe-style carons on ď ľ ť Ľ. The Lab adds Slovak, Croatian, and Slovenian samples to the existing ten-language inventory. New accents use ACC/MARK and the existing fixed-gap, lighter-mark attachment. ģ uses a turned comma above; new ogoneks receive their own punctuation clearance without changing approved pairs.

Step 2 (0.38) adds the Russian alphabet and common Slavic Cyrillic set (U+0400–045F plus Ukrainian Ґ/ґ), with pinned Russian, Ukrainian, Belarusian, Bulgarian, Serbian and Macedonian samples. It also adds the modern monotonic Greek alphabet, tonos, diaeresis, combined dialytika/tonos, final sigma and Greek punctuation. Identical shapes reuse existing glyph construction; new shapes use the same stroke, contrast, roundness and italic rules. Cyrillic italic б в г д и п т uses cursive forms. New script classes receive automatic pair measurements and separate clearance exceptions; existing Latin pairs remain unchanged. Historical Bulgarian yat/yus and Greek polytonic auxiliary letters are outside this batch.

Step 3 (0.39) completes Vietnamese: all U+1EA0–1EF9 letters and the four Ơ/ơ/Ư/ư horn bases are encoded. Eighty-eight missing letters use ACC recipes built from existing vowels, circumflex/breve parents and lighter tone marks. Canonical composition handles both stacked accents and structural vowel plus tone input. Horn and tone clearance uses literal pairs for new glyphs; prior letter outlines, advances, weights and spacing remain unchanged. The Lab adds a pinned Vietnamese inventory/sample and every new letter. The vietnamese subset includes complete Latin text, all encoded vowels and their combining ingredients together. All 21 recorded locale inventories pass the declared scope; frequent capital stacks can use 1.5 line height after checking the app.

phase4.json records release metadata separately from unchanged settings.json. Run python verify_phase4.py to compare all old glyphs, advances, settings, spacing and shaping with the starting commit. FontBakery results are in the health check. Alphabet coverage does not certify sustained reading on a physical device.

Roadmap

• Phase 3: the device checklist and full-chapter reading test are ready. Run the checklist on physical iPhone, Android, Windows, and Mac devices and record the results; no device results are claimed yet.
• Multilingual expansion: Phase 1's Latin foundation and Phase 2's first ten Latin inventories are implemented. Phase 4 completes Latin Extended-A and supplies common Cyrillic, monotonic Greek and full Vietnamese. Further African open-vowel/stroke/tone letters and other scripts remain candidates. Device/native-reader evaluation remains separate
• Phase 5 completed: recorded name screening, selected ecosystem license, and embedded legal metadata in all styles and subsets. Name screening is preliminary; no registered-trademark claim is made.
• Later: a separate SEIHouse display font

License

SEIHouse Sans is owned by SEIHouse Productions LLC, Ohio, and uses the SEIHouse Sans Ecosystem License 1.0. Ecosystem users may install, customize, and use it for personal or commercial creative work, including PDF/EPUB document embedding, without fees or user-count limits. Authorized SEIHouse apps, websites, SEA/SEN tools, and official font deliveries may distribute it. Independent app/product embedding and standalone font redistribution require written permission. See the full license for collaboration, notices, and scope.

The font's current family name is SEIHouse Sans. Existing SEIReader-* asset paths, @seihouse/seireader, and its styles.css / font-family: "SEIReader" alias continue working. New integrations can use sans.css / "SEIHouse Sans". The license is shipped in every runtime ZIP and npm package. Font metadata carries the owner, studio designer credit, project/license URLs, and license summary. fsType=0 permits installable document embedding subject to this EULA; it does not grant unrestricted standalone redistribution. Vendor ID SEIH is a project identifier; no Microsoft vendor registration or SEIHouse Sans trademark registration is claimed. The name screen records its limited scope.

The separate Display prototype and archived earlier releases are outside this license's scope and retain their existing permissions. The third-party reference fonts in references/ are separately licensed under the SIL Open Font License; their license texts are included in each family directory. The pinned CLDR locale data is separately covered by the Unicode license.