Skip to content

Bogen 0.4.0

Choose a tag to compare

@marselnenaj marselnenaj released this 18 Nov 15:44
· 3 commits to main since this release

I made Bogen something you can talk to: call your agent and it answers in a sentence or two while it works on your file. Files open as tabs, a brand opens on a real brand book, and pictures come from your own Codex CLI.

Call your agent

  • Talk with your agent instead of typing. The phone button in the chat head (or "Call your agent" in the command palette) puts the call on top of the chat as one calm strip: a status line for who has the line and for how long, a quiet line for what the agent is doing, and Mute, More and End. Under it is the last thing said, in a space that is kept, so the chat does not jump; the whole transcript opens from there. The voice, how the call hears, hold to talk, moving with the file, sounds, agent calls and what the call remembers are in Call settings, from More. The agent's answer, its tools and the details stay in the chat underneath.
  • A conversation, not a reading. The voice answers in a sentence or two and never reads the agent's text out: "On it." when your request is in, quiet while the agent works (one soft sound if it takes a while), and a short spoken result at the end, such as "The headline is bigger now. Does that work?"
  • It lets you finish. A pause mid-sentence ("Make the headline bigger and ...") no longer ends your turn, and a "mhm" or "yes" while the voice speaks only lowers it for a moment. Real words over the voice stop it at once; "stop" stops the run, "undo that" takes it back.
  • It answers what you ask, and asks back when you are vague. "Which design are we talking about?", "What's on this page?", "Are you done?" and "What did you just do?" are answered by the voice from what is on screen and from the agent's last answer, without starting a run. "Change something" gets one short question back, about the agent's last offer when there is one. When the agent ends with a question or an offer, the voice relays it, and your "yes" goes to the agent. The voice is told never to claim a change the agent did not make.
  • Who speaks is your choice. A signed-in Claude Code on your plan (Opus at low effort, no tools) speaks by default, because it retells the agent most faithfully. Without it, a small open model on your computer (Qwen3 4B on a GPU, Qwen3 1.7B otherwise, Apache-2.0, downloaded on request and checked by SHA-256), or plain rules.
  • A natural English voice. English is now spoken by tiny-voca (Apache-2.0, trained on the CC BY 4.0 tiny-voca dataset), a 21 MB model with two speakers, F2 and M2, that reads a sentence in about 30 ms on the CPU. Its phonemizer, dizione, is ported to TypeScript, so nothing native is needed. The voice is chosen in the call bar; LJ (Piper) stays selectable. Calls speak English.
  • Heard on your graphics card if you like. With a GPU, Whisper large-v3-turbo (MIT, about 574 MB) stays loaded during the call and writes a sentence down in about 0.2 s, primed with the names of your file, page and artboards; without one, Whisper base runs through Voxtype. Everything runs on your computer and is checked by SHA-256 on download.
  • Every call is kept with the chat. What the voice heard and said, how it decided and how long it took, so a call can be looked at afterwards.
  • Say more while it works. What you add while the agent is working ("more modern, full screen", "use Fable as an advisor") reaches it at once: the run stops where it is, what it changed stays, and it carries on with your words as the next message. Only a task you mark for later ("after that…") waits for the run to end. The voice says what it understood each time instead of the same line, and never the same acknowledgement twice in a row.
  • "Keep me updated" gives a one-sentence update about every minute while the agent works, from what it is doing; "be quiet" stops them.
  • No minutes of silence. In a long run the voice says one short line about every 90 seconds from the agent's latest steps, never the same news twice; "be quiet" stops it.
  • Small requests run faster. When you have not chosen a model, Bogen picks one by the size of the request: Sonnet at medium effort for a small change ("make the headline bigger"), Sonnet at low effort for a question, and the default model for new screens, a redesign or anything unclear. The chat shows the choice next to the agent's name ("Sonnet · small change"). A model or effort you chose in the chat, in run settings or in the call always wins. Agents also check their edits with the layout map or design_lint, look at a picture only first and last, and run design_verify once at the end.
  • No more "Thank you." from nobody. Before anything is written down, a voice activity model (Silero VAD, MIT, under 1 MB, downloaded with listening) checks that there is speech at all, so silence, room noise, typing, a cough or a door never reach Whisper. Lines Whisper is known to make up on noise ("Thank you.", "Thanks for watching", "Untertitel im Auftrag des ZDF") are dropped when little speech was behind them, and the call's own voice heard back through the speakers is recognised and ignored. The call record shows which engine heard each line, and keeps what was taken for noise and why.
  • The call hosts the conversation; the agent gets a brief. What you say for one task is collected until you are quiet (about 2.5 s, or say "go"), then sent as a brief: a goal and conditions in plain words, your own words marked as speech recognition, your corrections ("no, I said not happy" replaces what it corrects), and what else you said to the voice. When a request is unclear or would throw away a lot ("from scratch"), the voice asks before it starts. Only work reaches the agent: thanks are not answered at all, remarks ("that's awesome", "I'm waiting") are only listened to, "what?" and "is it still working?" are answered by the voice, and nothing of that ever stops a run. Changes you say while it works go together after about 3 s of quiet; tasks for after are kept as a list, each sent on its own.
  • The agent keeps what it knew. A call run keeps its Claude Code session: a change ends the turn the way Ctrl+C does, and the next run continues the same session, so the agent does not read the whole file again. The agent knows it is in a call: it never asks you to paste anything, and ends its answer with a short line for the voice, which is spoken as it is. A pause mid-sentence now waits up to 2.8 s, never longer than 12 s for one turn. Replayed against a real call of 16 Nov: 13 messages reached the agent instead of 105, 3 lines that were not requests instead of 89, 5 interruptions of a run instead of 52 (none for a remark instead of 44), and every status question answered.
  • Hears you better, on any computer. NVIDIA Parakeet TDT 0.6B v3 (CC BY 4.0, through sherpa-onnx, Apache-2.0) is now how a call hears: 25 European languages in one 671 MB model on the CPU, about 0.3 s a sentence, with the names on screen as hotwords. On real voices it made about half Whisper turbo's word errors, and lost no sentence to a made-up "Thank you." (Whisper turbo lost 6 of 100). Whisper on the graphics card stays a choice in Call settings.
  • A question changes nothing. Asking the agent to check, review or say what it thinks ("What do you think?", "How does it look?") no longer ends in changes you did not ask for: it gives its view and offers changes instead. Questions written down without their question mark, as speech recognition does, count as questions too.
  • The call goes where you go. Leaving the file, going home or opening another file keeps the call, its microphone and its voice. In its own file the call is a card under the chat's header; everywhere else a bar at the foot of the content column shows the same: a ring and a glyph for who has the line (blue is you, amber is the agent, red only End and problems), a title that says what happens, what the agent does and in which file, the clock, and Mute, More and End in the same places in every state. What you said is in the card's transcript, from More; call settings are in More too. Hanging up leaves a short note with Call again. At phone width the bar spans the screen with 44 px targets.
  • It talks like a person, not a phone menu. With Claude Code as the talking layer, one model now decides and words every turn: it answers small talk and "hello?" in its own words, asks back when a request is unclear, hands only real work to the agent (in its own words, never reading yours back), changes or stops the running task, and moves the call to another file when you name one. What it says about the agent comes from the server's record, so "is it working?" is right also for runs started from the chat. The fixed lines are only a fallback when it does not answer in time. Measured on 11 messy calls, among them a real one of 17 Nov: turns that went to the agent although they were chat, got no answer or a wrong state went from 9 of 58 to 1 (a question the echo filter took for the voice), an Opus judge scored the calls 6.5 of 10 instead of 4.1, and the first word comes after about 1.3 s (1.8 s at most for nine in ten turns).
  • It knows when to keep quiet, and starts sooner. The voice says nothing it already told you during a run: a line that mostly repeats one of its last three is cut to its first part, or left out while the agent works and nobody asked. "Alright", "got it" or "you're amazing" while it works get no answer, and "you don't need to reply to every thanks" is kept for later calls. Half sentences ("Could you...") are let hang, with at most one "Mm?" a minute. A clear request gets a short "Okay." the moment you stop, and goes to the agent with the brief the voice wrote on its decision line, about 1.2 s after your last word instead of 5.5 s. The voice speaks from the first comma, the voice heard back is told from you by when it began (a question is never dropped), talking over it stops it after half a second, and a line you talked over is marked as not heard. Files are named by their short name, a stop is not followed by a result, and one question is asked at a time. On 14 scripted calls, among them a real one of 17 Nov, fixed lines went from 15 to 0 and no turn failed a check (3 before).
  • Call sounds. Soft tones of its own, made in the browser, for dialling, connected, ended, mute and unmute, an incoming call and a failed one. Never over the voice, and off with one switch in Call settings.
  • Your agent can call you. When a run you asked for (in the chat, a comment or from your phone) ends on a question, Bogen rings for 30 seconds with the question: Answer opens the call in that file and the voice asks it first, Reply in writing opens the chat or thread, Decline leaves it there. A missed call stays as a note. Off until you turn on "Agents may call you" in Call settings; never in your quiet hours, never while another call rings, never more than once every two minutes, and never twice from one file within ten minutes. A paired phone gets a notification that names the agent and the file, never the question; a desktop browser in the background shows one when you allow it.

The call remembers

  • A long call keeps short notes on itself, written every few turns while nobody speaks: the voice works from them and the newest lines instead of only the last lines, and a later call about the file starts from them.
  • Open another file during a call and the call goes there too, with a soft tone and the bar naming the file for a few seconds. A run still going in the old file finishes there. It can be turned off in the call's settings.
  • Wishes about how the voice talks ("keep it short", "from now on, always ask before you start", "call me Sam") are kept for every later call, in call-preferences.json in the data folder. The call's settings list them, and each can be forgotten.
  • "Use the Second Shelf brand" finds the brand by its name and hands the agent its id; with two close names the voice asks which, and an unknown one is said so. Nothing is created.
  • A call that ran requests ends with a note: the files it changed and the requests in each, stopped ones marked. It stays until closed and is kept with each file's chat.

Files as tabs

  • Several files open in one window as tabs above the toolbar, with Home first. Opening a file from Home, a link or bogen open adds its tab; the tabs come back after a reload.
  • Switching back to a tab is instant: its file stays loaded for a while behind the tab, with its selection, panels, page, view and undo steps, and keeps syncing. A tab left long enough, or more than two behind, goes to sleep and loads its file again when opened; a big captured site sleeps after a minute.
  • A mark on a tab says what happens in that file: an agent at work, the call, no connection, edits waiting to sync, someone else in it.
  • Ctrl+Tab and Ctrl+Shift+Tab (or Ctrl+Page Down and Page Up) switch tabs, Ctrl+1 is Home, Ctrl+2 to 8 a tab by place, Ctrl+9 the last, Ctrl+W closes, Ctrl+Shift+Page Up and Page Down move the open tab. Middle click closes, tabs reorder by dragging, arrow keys and Delete work in the strip. With more tabs than fit, All tabs lists them.
  • On the phone the open files are a sheet behind a button in the file's top bar.
  • A file is open in one window at a time. Opened in a second window of the same browser, that window says "This file is open in another window" instead of showing a second editor, and asks the other window to come forward; Use this window moves the file over (the other window goes Home), Close goes back. bogen open, the Omarchy launcher and the Bogen bar use the window that has the file, or one in view, before a new window opens. On a Mac the app brings that window forward. A phone or another computer is not affected.
  • Two open files with the same name read differently in the tabs: by folder, as a kept copy, or by the start of the file id. Every address of a file (with a page, a layer, a thread, or the id in capitals) is the same tab.

Brand book, Brands and the prompt box

  • A brand opens on its brand book: landscape guideline pages made from the brand's data, on its own paper and ink, with a cover in its primary lockup, a full-page opener in the brand's colour for each chapter, the logo with its clear zone, small sizes and misuse, colour as columns sized by their share with HEX, RGB and OKLCH and measured contrast, type specimens and named steps with real copy, space, voice, imagery with do and do not, rules, tokens grouped as core, semantic and component, and applications drawn from the brand (business card, email signature, letter, social post, web page) next to its own work cropped so it can be read. Approval state and what is still to add are in the colophon, not on every value. The PDF has one page per page, and on a phone the pages reflow. brand_sheet shows the same pages to agents.
  • The brand page's sections are grouped into Edit, Write and Use, and each has its own address. Across files is now Files using it, and the chapters are under Write.
  • Files and Brands are places on Home. Brands lists every brand and takes one from a website address, with progress, partial results and a clean failure; what it finds arrives as suggestions to review.
  • A prompt box at the top of Home makes one file of a kind (website, app interface, Omarchy theme), with a brand, in the folder you are in, and starts your agent there. A link in a prompt offers its brand.
  • A brand card or a file card on Home opens from anywhere on it, with a focus ring around the whole card.

A designed edition of the brand book

  • Make a designed edition, above a brand's book, has an agent art-direct the guidelines for that brand: a new file "Name · Brand book" with the generated book as a locked reference, 10 to 14 landscape pages with one idea each, applications from the brand's own world and its quirks on show, nothing invented. It runs with your chat's agent and model, on your plan, shows in the file's chat and the activity dock, and is reviewed once by design_critique, a fresh reviewer on another model briefed as studio guidelines, whose findings the agent fixes (a loop of up to four rounds only when asked for). design_critique itself now answers within 45 seconds and hands a review that is still going to the next call, so it works under the 60-second limit of Claude Code's MCP calls. The brand page then links to it, says which parts of the brand changed since, and refreshes only those pages. Agents start it with brand_book_design; the method is a new design playbook, brand guidelines books, with the book's arc and applications by kind of business as compositions.

Generate image

  • Generate sits next to every Upload image: in the Image section, the image fill editor and a shader's picture, and as Generate image in the command palette. Describe the picture, pick a shape (guessed from the layer) and up to four takes, then choose one from a sheet with the mouse or the arrow keys; it goes into the layer with the description as its alt.
  • It uses the Codex CLI on this computer with your own sign-in, about a minute per picture, two at a time. Each runs in an empty temporary folder in Codex's sandbox with its shell turned off, and the picture is stored as a WebP like an upload. A take that fails says why and can be made again; Stop ends the rest.
  • Without the Codex CLI, or signed out, the dialog says how to set it up and checks again on request. Nothing else changes.
  • Agents get the same as image_generate, which can fill an image layer or a frame's background.

Review like a designer

  • An independent reviewer inside Bogen. Review like a designer (in Design checks, the command palette, and as design_critique for agents) renders the screens, runs the rule checks, and hands the pictures, the brand's rules and a studio brief to a fresh Claude Code session on your plan. It answers with a score out of 10, scores for state clarity, consistency, hierarchy, copy, layout and brand, every objective defect with the evidence it saw and a fix, taste notes, and what to keep. A defect points at its layer and can be pinned as a comment.
  • Never the maker's own eyes. Every review is a new session that sees only the pictures, the brand and the brief: not the agent's session, its reasoning or earlier scores. After a fix it is told only which findings the fix took on, and checks in the pictures whether each really is fixed. It is a different model from the one that made the work (Fable reviews Opus's work and Opus reviews Fable's), and a review without evidence for a finding moves it to taste.
  • Until it is right. An agent can loop: fix the defects, call again, and a new reviewer looks, until no objective defect is left and the score meets the target, or the rounds run out. Each round is kept on the file, so you see the history.
  • Waiting you can watch. While the reviewer looks, the dialog pins your own screens to a wall, as the pictures the reviewer got, with a reading light walking along them and pencil marks where it has been. The stages are the real ones (rule checks, pictures, the reviewer looking, writing up, with the findings counted as they are written), the clock is real, and there is no made-up percentage. With reduced motion the wall is still.
  • A crit sheet. The score fills its ring once, the six dimensions sit in one row, and each defect shows where it is on your screen, with Fix, Pin and Show layer. Taste notes and what to keep follow; the rounds of a loop are a small line of scores.
  • Fix at once. Fix on a finding, or on all of them, starts your agent right away with the chat's agent and model and a brief built from the findings: no draft to send. It shows in the chat and can be stopped and undone as any run. With "Then review again until 9", a new reviewer looks after each fix, until no defect is left and the score is 9, or after three rounds.
  • Close it any time. X, Escape and a click beside the dialog always close it. A review, a fix or pictures still being made go on as a pill in the dock at the foot of the window, where the call bar is, with Open and Stop; three or more fold into one. A brand being taken from a website shows there too when you leave Home.
  • The rules catch what a reader catches. design_lint now flags a double negative, an unclosed quotation, the danger colour on a calm state such as "working", one label spelled two ways across screens, and text that breaks what the brand says about capitals, italic headings, the heading family or shadows. Each is a hard finding, and goes to the reviewer as input.

App screens that hold up

  • Phone screens are measured. design_lint knows an iPhone or Android artboard and checks what reviewers kept finding on generated app screens: text and buttons under the status bar or in the home indicator strip, real controls under 24px, edges, gaps and paddings that slip by 1 to 3px, and a button label off its centre. Each is a hard finding. Targets under 44pt or 48dp, a tab bar off its size, a large empty band, small sentences, shapes drawn as buttons and a screen without a platform kit are advice, with the numbers and why.
  • Start from the platform kit. The iOS kit now has the parts app screens need at iOS sizes: status bar, home indicator, navigation bar with a large title, inset grouped list sections and rows, tab bar and tab items, sheet, empty state and full-width buttons, with every text a prop. The Material 3 kit gets its status and gesture bars. guide_get apps tells agents how to build a screen from them, with a spacing system, a lead for each screen type and the platform's numbers.
  • Done means reviewed. An app screen counts as done when design_lint has no hard finding and the independent reviewer's loop scored it 8 or more, with what it still lists reported; design_verify shows where that stands, and the reviewer now also checks spacing, hierarchy and the platform on phone screens.

Brand images

  • image_treatment gives a photo the brand's own look, on this computer and the same every time: halftone (dots, squares, rounded cells or lines), mosaic pixels that can dissolve toward a side, ordered or error-diffusion dither, duotone and posterize, in the brand's inks. With inkBy region the subject and the backdrop each get their own ink. Halftone, mosaic and dither come out as an SVG whose fills follow the brand tokens; the others as a PNG.
  • A treatment can be saved in the brand as a preset, with an example before and after. The brand page shows presets under Imagery, the brand guide and sheet have an Image treatment section, and agents in a linked file use the preset for photos.
  • image_vectorize with palette brand traces to the brand's colours only.
  • image_search finds free photos on Openverse and Wikimedia Commons: only CC0, public domain, CC BY and CC BY-SA, each with author, licence and source. asset_upload from_search imports one and keeps its credit with the file, as the provider states it; image_credits lists what a file owes, with text for a footer or a closing slide.
  • shader_params lists every shader effect's parameters with defaults, choices and ranges. An attribute an effect does not take is now refused with the list, instead of doing nothing. An effect's own type, such as a dither's 8x8, is set as variant; choosing such a preset in the inspector no longer replaces the effect.

Brand applications

  • Mockups: mockup_place puts an artboard on one of nine scenes drawn in Bogen (poster on a wall, billboard, tote bag, badge, hanging sign, packet, cap, phone, laptop) or on a photo of your own, by four corners or a mesh for cloth. The design stays live, so a changed poster changes every mockup, and the surface's light falls on the print. In the editor, Place in a mockup in the command palette, and a Mockup section to swap the scene, blend, light and fit.
  • Graphic devices: pattern_set draws a brand's motif as cells that fill by a gradient, noise, a figure (wave, mountain, tower, cloud), a bar chart, an image or a word, in the brand's tokens, as vector. A brand keeps its devices as presets in a new Graphic devices section. The inspector tunes a pattern without an agent.
  • text_outline turns a text layer into outlines from the font the canvas uses, with its kerning and tracking: the way to build a wordmark.
  • brand_presentation presents a brand as an agency reveal deck (26 slides from cover to close) and a guidelines book (clear space, misuse made from the real logo, colour values with CMYK and every contrast pair, type scale, grid, device), from the brand and a story the agent writes. Slides whose story is missing are left out and named. presentation_check says what reads as finished and what reads as generated.
  • A file linked to a brand that is still all suggestions now gets the brand's tokens, marked provisional, and brand_check counts them.
  • Posts, stories and posters take one of seven forms (photo with the line in a safe area, split, the one number drawn in the device, quote, device as figure, type alone, product details), and no two neighbours share one; presentation_check names repeats.
  • A brand with an image treatment gives every photo in its deck and site that look by default.
  • The system wall is its own board: every application in its own ratio and the device at favicon, card and full-bleed size, each tile checked against the brand.
  • brand_site builds the brand as a website to publish: chapters with version and changes, colours with their measured share and every contrast pair, tokens to copy for CSS, JSON, Tailwind, Swift and Kotlin, a type tester, logo construction to switch on and off, the real applications and motion from the file.
  • Mockups keep the whole design in view when its shape differs from the surface, and a selected mockup has four corner handles on the canvas.

The editor

The editor's own interface was captured into Bogen and its design checks were worked down from 280 findings to 0. These changes carry that pass over into the editor:

  • Clearer toolbar: tools in groups with separators, and the device sync state next to the file name, written out instead of cut short. On the right, Preview has a label and Ask agent (Ctrl K) is the main button.
  • What changed while the file was closed now appears in the recent-changes strip, with Review changes, instead of in a separate notice.
  • With nothing selected, the inspector shows the page (artboards, sections, open comments) and its design checks, with Run checks and Ask agent to fix.
  • The left panel tabs show plain words, and Data is named. Page times are shorter, the page list is no longer fixed at 148px, and the canvas map shows Sections.
  • Readable at a glance: interface text is at least 12px, buttons and toggles are at least 24px, the sync pill meets contrast, lane labels no longer fade across a section border, and section descriptions wrap. The spacing follows one scale of 2, 4, 8, 12, 16 and 24px.

Faster saves

  • Big files save as a log of their edits and are folded into the full file when idle, on unload and on a clean stop: on a 72 MB file, 20 small edits wrote 5.3 MB instead of 179 MB, and the longest pause fell from 505 ms to 24 ms. The full file stays in the format older builds read.
  • An edit clears only the cached layers it touched; font links and shader parameters are cached until the file changes.
  • Recent edits moved out of index.json into each file's history, so the file list is small and only rewritten when it changes. Tool results are compact JSON.

Security

  • Every listener refuses paths spelled another way (empty or dot segments, backslashes, encoded slashes), which closed a way around the tailnet and phone route blocks. Peer administration is only allowed from this computer.
  • The headless browser reaches the web through a guard in the server that resolves each address once and refuses private ones, which closes DNS rebinding, and WebSockets go through it too. Local captures may only fetch from this computer.
  • batch checks the branch, revision and critique rules like a single write; image decoding and received commits have size limits; the editor has a strict content security policy; every dependency is pinned and npm audit is clean.
  • Thumbnails are written atomically and temp files left by a crash are removed on start.

Fixes

  • bogen mcp, bogen stop and the other commands keep to the data folder BOGEN_HOME names: they use the server that folder records, and refuse a server that serves another folder instead of writing into it. Before, a test with its own folder could reach the everyday server on the default port.
  • A tab opened after the window fetched the file names (a link, the call moving to another file, a new file) showed "Untitled file" until the next change to the file list; it now gets its name at once.
  • Switching tabs is Ctrl+Tab and Ctrl+Page Up/Down on a Mac too, where Cmd+Tab is the system's app switcher; Cmd+1 to 9 and Cmd+W stay.
  • The brand book sets a long type scale on one or two pages, each size with words from the brand instead of its name.
  • The brand book follows the brand's own rules (capitals only, italic headlines, no shadows, no dashed lines, no screenshots of other brands), and scripts/check-brand-book-render.ts fails a page that breaks one. Its applications carry only the brand's own words: no invented names, grey placeholder lines, agency navigation or window chrome. A brand with little written down gets a short book without openers, and logos drawn with an empty margin are trimmed to the mark.
  • The brand book fits its pages: nothing runs past a page or into its folio, the cover and the brand idea lead with a short statement instead of a cut sentence, the core colours are large and the rest sit in one table grouped by job, only the contrast pairs that matter are shown, working notes from the data (file paths, sources, notes to agents) stay in the Markdown export, and a German brand gets a German book. I check every page of every brand on my machine in a browser for this (scripts/check-brand-book-render.ts).
  • The brand book keeps its own rules in its pieces too: a logo only sits on a ground it is made for, so the business card no longer shows the mark on the colour its misuse page crosses out; screenshots stay whole and portraits grey when the brand says so; display lines take the brand's width; headings stay in its heading face. It closes every quotation it cuts, never says no twice ("Never Do not"), marks unconfirmed rules as proposals, draws corners, shadows and gradients next to the rules about them, and shows the brand's example artboards from its linked files, sharp. The render check now also fails a logo on the wrong ground, a doubled rule under the running head, a double negative, an open quotation, a heading in the body face and an enlarged picture.
  • The brand book holds its own copy to the brand's voice: a headline says one thing when the brand asks for that, a brand that speaks as "I" is not shown saying "we", and an example that breaks the brand's rules is left out and counted in the colophon. Every column sits on one twelve-column grid, every page has the same running head, quotes and apostrophes are typographic and hex values in capitals. The colophon counts the chapters the contents lists. Openers carry their chapter's own thing (the mark, the colours, the letters, the spacing, a picture, the tokens, the brand's work), the brand's real screens lead the applications, sized to show whole or cut where no line is cut, and a picture whose text would be too small to read is left out. scripts/check-brand-book-render.ts now also fails boxes past the margins, off-grid columns, a second head system, straight quotes, lower-case hex, ornamental quote marks, unreadable screenshots, captions that repeat their heading, list columns without their heading, samples that fail the brand's voice rules and a colophon that miscounts.