The One-Page Test

The decision it forces

Is this output actually saying something, or is the formatting doing the work?

A model hands back seven pages. Headings, bullets, bold phrases, a summary box, a section called Next Steps. It looks finished, and that is the problem, because looking finished is the cheapest thing a model produces. Lee Rodrigues, on Episode 1:

And what I found is I was getting lost in the sauce with all the formatting and all these fields and everything. And you can't remember how much of this is me, how much did it come up with, how much relates to what I do.

To run it, copy the prompt below into any model you already use.

The move

Lee's, from before there was a model to run it on. When he trained designers, the first deliverable was a one-page content outline that defined what everyone was talking about. Faced with the model's output, he asked for the same thing:

said, hey, this is a beautiful seven-page document that I'm already lost in.

Back to the one page, without the polish, and the structure let him see where the model was making things up. The strip list he gave on air is the whole instruction: subtract the formatting, subtract the bars, subtract the just-in-case context. Sam's gloss in the same conversation was to remove the flashy sparkles and ask what the actual thing is that can't be bluffed.

The format, when Sam asked, was Lee's two words: "T X T or Mark Down". No document format, no styling, nothing a renderer can make look better than it is.

The fields

The re-render is not the deliverable. The diff is. Three lists come out of the compression, and two numbers.

There is no worked example on this page. A filled test is one document of ours with its unsourced claims listed, and publishing it would say more about the document than the method.

The verdict rule

Read the count, not the outline. evaporated is the number that lands. Seven pages that compress to eleven lines is not a summary, it is a measurement of how much of the document was formatting. unsourced is the risk: the claims that read as fact and came from nowhere in what we gave it. Lee's version of the second list is the reason the test exists: "It's finding things that were not a part of my content that it's bringing in from somewhere else, and I can just trim to subtract what isn't a part of it".

Run it with any model

Nothing to install and no account. Paste the output and, if there was one, the source material it was supposed to come from. The model compresses and classifies; it may not add.

Run the One-Page Test on a document I am about to paste. The document was
produced with AI assistance. I want to know what it is actually saying once the
formatting is gone, and which of its claims rest on nothing.

If I also paste the source material the document was supposed to be built from,
use it for step 4. If I do not, say so and treat every claim as unsourced except
ones that are self-evidently definitional.

DO, IN ORDER
1. Count the words in the document and record source_length_words.
2. Re-render it as a one-page content outline in plain text. No headings styled
   as headings, no bold, no bullets deeper than one level, no tables, no summary
   box, no section that restates another section. If it does not fit on one
   page, it is not compressed enough; cut restatement first, then hedges, then
   transitions. Every line must be a claim, an instruction, or a fact.
3. kept: list every claim that survived, one per line, in the words the document
   used.
4. evaporated: list what you removed, grouped as transitions, hedges,
   restatement, or just-in-case context, with a count of each. Give a total in
   words.
5. unsourced: of the claims in kept, list every one that does not appear in, and
   cannot be derived from, the source material I gave you. If I gave you none,
   list them all and say why.

RULES YOU DO NOT BREAK
1. Do not add a claim that was not in the document. The outline may only be
   shorter than the source, never different from it.
2. Do not soften the unsourced list. A claim with no source stays on that list
   even if you believe it is true, even if it is common knowledge, even if it is
   obviously correct. Whether it is true is not the question; whether it came
   from somewhere is.
3. Do not rewrite for quality. Do not improve a sentence. Do not fix a fact. If
   the document is wrong, the outline is wrong in the same way and the unsourced
   list is where that shows.
4. Do not set the verdict. Leave it as "unresolved". What to do about the result
   is mine.
5. Do not tell me the document is good or bad, thorough or thin. Give me the
   outline and the three lists and stop.

OUTPUT
Valid JSON matching the schema one-page-test/v0.1 as published at
https://sigsub.show/takeaways/one-page-test/, then the outline as plain text on
its own, then one line: source words, outline lines, evaporated words, unsourced
count.

What to check in the answer

It may compress, strip, classify, and name what disappeared. That is all arithmetic on text it was given.

It may never add a claim that was not in the source, and it may never soften the unsourced list. The moment a model decides a claim is fine because it happens to be true, the test has stopped measuring where the document came from and started measuring what the model believes, which is the thing we were checking in the first place.

The part that's not optional

Read the unsourced list before reading the outline. The outline will look reasonable; it always does, that is what compression does. The unsourced list is the part that tells us whether the reasonable-looking page can be sent to anyone.

Then, if the document is going out under our name, write the missing sources in or cut the claims. Lee's warning on the episode is what happens otherwise: "you end up sharing it with a client or something, and they say, Where did this come from?"

{
  "schema": "one-page-test/v0.1",
  "source_length_words": 0,
  "outline": "plain text, no formatting",
  "kept": ["claims that survived compression"],
  "evaporated": ["text that carried no claim"],
  "unsourced": ["claims that survived but rest on nothing in the source"],
  "verdict": "unresolved"
}