Sa74/break-the-template

Break the Template

You were asked to make something engaging, and you are about to produce a document: headings, paragraphs, a figure with a caption, repeat. This skill exists because that instinct is strong, invisible to you, and almost always wrong for this kind of request. It was distilled from a real session where it took an author four rejected attempts to escape their own template — the lessons below are what finally worked.

The trap you are in (read this first)

Three biases produce template-shaped work, and none of them feel like biases from the inside:

1. Inherited structure feels immovable. The existing page layout, the essay form, the slide deck grid — you treat these as the laws of physics when they are just the previous decision. Nobody ordered a prose column.
2. Decoration feels like reimagination. Adding animation to a diagram, theming sections, renaming chapters "tracks" — these change the skin while the reader's experience (scroll, read paragraph, look at box, repeat) stays identical. If the consumption loop is unchanged, you changed nothing.
3. You optimize against the failure you can measure. Builds break loudly; blandness is silent. So you converge toward safe, valid, forgettable. Treat "a reader would not screenshot any part of this" as a build failure.

The tell that you're trapped: your plan describes sections before it describes what the audience does.

The method

1. Name the template you're about to follow

Say it explicitly: "the default here is a scrolling article with section headings" / "the default is a 12-slide deck of bullet points". You cannot break a structure you haven't admitted you're using. Then write the kill-list: the elements you are deleting, not restyling (e.g. "no paragraphs over 2 sentences", "no figure captions", "no section headings"). If the kill-list is empty, you are decorating.

Beware the false escapes — structures that feel creative because they animate, but are templates all the same: scrollytelling chapters, hero-sections with count-up stats, card grids, vertical timelines, tabbed panels. In benchmarked head-to-heads these are exactly what capable models produce without this skill. If your "new" structure appears on a component-library homepage, it's a template with confidence.

2. Find the medium's native superpower

Ask: what can THIS medium do that the template ignores? A browser can make sound, respond to clicks, know where you've scrolled, run a simulation. A deck can control pacing and reveal. A README renders inside a developer's moment of evaluation. The strongest experiences let the audience do or feel the subject, not read about it: a piece about typography should let you set the type; a post about latency should make you wait and feel it; a doc about a maze-solver should let you race it.

3. Find the conceit — one governing metaphor, fully committed

A conceit is a world the content lives in, drawn FROM the subject itself: a making-of told as a workshop you walk through; a migration story told as a moving-day; an API doc told as a restaurant menu. Two rules make a conceit work:

• It must come from the subject. A photo app's story belongs in a darkroom; a random "space theme" is a gimmick. If the metaphor could be stapled onto any topic, it's wallpaper, not a conceit.
• Total commitment. Translate EVERY element into the conceit's vocabulary — headings become signage, the org chart becomes a parody of the product's own UI, trade secrets become a safe with torn-off cards, the FAQ becomes the complaints box. The moment you fall back to a plain section, the spell breaks. Half a conceit reads worse than none.

4. Decide what the audience does, and budget their time

Write one sentence: "The reader spends N minutes doing X." (e.g. "3 minutes walking through a room, pressing two things.") Then enforce a copy budget — the sticky-note rule: no unit of text longer than two sentences, total words counted and capped. Long-form prose is what the template wants; resist it everywhere. Information you'd have written as paragraphs becomes labels, speech bubbles, captions on artifacts, marginalia.

Compression must cost zero facts. Before designing anything, write a fact manifest: every number, name, date, and claim in the source material, ranked by what the audience came for (a recruiter reads an About page for seniority and proof; a team reads a quarter recap for the numbers and the honest miss — those facts are load-bearing and uncuttable). After building, walk the manifest and confirm each fact survived — a unit change (1.9s → 1900ms) is fine; an omission is a defect, however small it seems.

And the mirror rule: invent nothing. When a piece turns playful, the strong temptation is to fabricate charming specifics — a fake uptime counter reading "2,191 days", a made-up "27 layers", a derived statistic whose math nobody checked. These read as facts and poison trust in all the real ones. Make jokes with form (a blinking sign, a sleeping robot, a torn card), never with data; if a gag seems to need a number that isn't in the source, the gag is wrong, not the source. Creativity spends on form; the facts are not its budget.

5. Pitch before you build

Generate 2–3 genuinely distinct concepts. Distinct means different consumption models — explore vs. scroll vs. play — not three skins of the same page. For each: the conceit, what the audience does, what gets deleted, one vivid detail that proves you've imagined it (the detail sells the pitch: "the office plant has been dead since week one"). Recommend one. If a human is present, let them pick or merge; if not, choose, and ship a short note with the deliverable naming the concept and the alternatives you rejected.

6. Build with craft rules

• Motion must teach or be physical. An animation earns its place by demonstrating a fact (a progress bar that really waits the full delay) or by behaving like a real object (a sticky note lifting). Looping decoration is noise. Same test for interactivity: pressing the button must cause the real phenomenon, not a simulation of it, wherever the medium allows.
• Constraints become features. Can't reveal trade secrets? Redact them theatrically — a safe, censor bars, torn cards — and say so. Limits played openly read as confidence; limits hidden read as gaps.
• Stay inside the design system. Use the host project's tokens, fonts, and themes. Breaking the template is structural rebellion, not visual vandalism — the wildest concept should still look native in light and dark mode.
• Floors that are not optional: respect prefers-reduced-motion (freeze everything), make sound strictly opt-in, keep semantic structure and alt text for what screen readers need.
• Verify with your eyes. Render every visual and look at it before calling it done. Layout collisions, clipped text, and wrong content survive in code that compiles.

When the feedback is "it's still the same"

Do not polish harder at the same level. Escalate one rung:

styling  →  structure  →  consumption model  →  medium
(colors,    (order,        (what the reader      (page makes sound,
 fonts,      sections,      DOES: explore,        becomes a game,
 motion)     density)       play, walk, press)    a comic, a room)

Most "make it engaging" requests are solved at rung 3 or 4. If you've been asked twice and you're still at rung 1–2, jump straight to 4.

Self-tests before you ship

• The manifest test: take the fact manifest and find each entry in the final piece, one by one. Anything missing goes back in before anything else gets polished. Then scan the piece for specifics (numbers, dates, named figures) that are NOT in the manifest — each one is either deleted or clearly framed as fiction.

• The textbook test: could this visual appear unchanged in a textbook? Then it explains, but it doesn't delight — push further.

• The topic-swap test: swap in a different subject; does the structure survive intact? Then it's a template wearing a costume.

• The ten-second test: what does the audience do in the first ten seconds? If the honest answer is "begin reading", reconsider the opening.

• The retelling test: can someone describe the piece in one sentence that makes a stranger smile? ("Her About page is a recipe card, and you can tear a tab off the bottom.")

Worked example (compressed)

A making-of page for a desktop app went through this arc: v1 — essay with diagrams (template). v2 — sections renamed, diagrams animated (decoration; rejected). v3 — a live demo of the app's central trick, bolted onto the same essay (right rung, wrong structure; rejected). v4 — the page became one continuous illustrated room the reader walks through, every fact an object they can look at or press, about 500 words in total. Accepted. The difference between v2 and v4 was not effort — it was rungs.