BIG Tools
Sign Up

Recipe: Paper House

A deterministic self-check — can you say your idea in one paragraph, under the word budget, with no markdown scaffolding?

DEPRECATED — ADR-038 (2026-04-27)

Dropped — making this an LLM-as-judge tool would have AI deciding when an idea is ready to share. Judgment is human work. The d.school principle survives as a reminder: keep concepts short, simple, easy to convey. The check is the user.

What this tool does

Paper House is the only BIG Tool with no LLM call. It's a three-rule check you can run against any piece of writing — especially the final pitch — to verify that the idea holds up without structure:

  1. Word budget — the text is ≤ N words (default 80).
  2. Single paragraph — no double line breaks. One continuous block.
  3. No headings — no # Title, no ## Section, no markdown scaffolding.

The premise: if you need headings, bullets, or a 400-word explanation for your idea to make sense, the idea isn't ready. A "paper house" is an idea that can stand on its own weight without structural props.

Students usually hit Paper House at the end of a progression, after Pitch. It's a self-check: can the X-for-Y plus elevator hold together as a single paragraph? If not, the pitch is hiding something.

The rules in plain English

  • Count words in the trimmed text (split on whitespace, drop empty tokens).
  • Split the text on double line breaks (\n\n). More than one non-empty chunk → not a single paragraph.
  • Scan for lines matching /^#{1,6}\s/ — any match → headings present.

All three must pass. If any fails, the writing isn't ready.

Example

Pass:

It's a deadman switch for AI. Every consequential action — a deploy, an approval,
a billing change — requires the operator to turn a physical dial or press a
physical button before the model executes. The friction is the point: speed
and opacity are what make AI systems dangerous, so we replace them with
deliberate, bodied participation. Early signals are good. Six-week pilot with
two teams is lined up for fall.

58 words. One paragraph. No headings. Passes.

Fail:

# The Idea

It's a deadman switch for AI.

## How it works

Every consequential action — a deploy, an approval, a billing change —
requires the operator to turn a physical dial or press a physical button
before the model executes.

Headings present (fail), two paragraphs (fail), well under budget (pass). Overall: fails.

Replicate it

You don't need to. This is a self-check. Read your paragraph, count words (or paste into a counter), and verify by eye that it's one block with no headings. An LLM isn't more accurate than reading it yourself.

If you want an LLM to judge it for you:

SYSTEM: Apply the Paper House check to the following text. Return exactly three lines:
WORDS: <count>
PARAGRAPHS: <count>
HEADINGS: <yes or no>

USER: <paste text>

But again — a wc command is better than an LLM call for this.

Tuning

There's nothing to tune. That's the point.

  • Word budget defaults to 80. Some students use 100 or 120 for longer-form pitches. Dropping below 60 is punishing — one paragraph isn't the same as one tweet.
  • Paragraph rule is strict — exactly 1 non-empty block. If you need two paragraphs, your idea is really two ideas. Decide which one is the pitch, put the other in a follow-up.
  • Headings rule is strict. If you think you need a heading, you're probably drafting a document, not a pitch. Use a different channel in Pitch instead.

Why these rules

The Paper House metaphor: an idea that only stands because you propped it up with structure isn't load-bearing. A heading is a prop. A bullet list is a prop. A 300-word paragraph that introduces three different ideas is three houses leaning on each other.

Pulling the props away reveals whether the idea has a center. If you can say it in 80 words, one paragraph, no headings — and someone who didn't see the progression understands it — it's a paper house that can stand.

If it can't, you're not ready to ship. Re-run an earlier stage. Re-open Mixtape. Go back to Reframe and sharpen the HMW. The right next move isn't adding structure; it's finding what's missing from the idea itself.

When to run it

  • After Pitch, on the elevator section. The X-for-Y line should be your opener; the elevator is the whole paragraph.
  • Before any external share — a tweet, an email, a pitch deck title card. If the core idea can't pass Paper House, it won't survive contact with a reader who doesn't owe you attention.
  • As a diagnostic. If you can't pass Paper House, you know where the progression is weak. Usually it's Mixtape (theme too generic) or Reframe (HMW too solutions-shaped).