The card path
If you own a notebook, you already own the good part.
What you come back with is the thing you have been going round and round on, decided. Not narrowed to three options and handed back to you: decided, in your own handwriting, in a notebook you already own, somewhere with nothing waiting on you.
You say one sentence. The session arrives as a card: one line naming what you are really deciding, one prompt per page, a few numbered handles where the AI has done work worth reacting to. You copy that part into the notebook by hand, take it away from the desk, and answer it in pen for as long as it takes. Photograph the pages afterwards and hand them back. What you wrote outranks anything proposed before you picked up the pen.
A notebook is what every sheet in this project labours to become: a mostly empty page where your pen is the highest-contrast thing on it and nothing printed argues with it. You have one already.
The sentence
Your AI is not allowed to ask you this, so you get to say it first.
Can you print, can you read the page: both questions are banned outright in the reference the skill loads before it designs anything. So the card path only ever answers something you said. Add this to whatever you were already asking for, or send it on its own after a PDF has turned up.
I can’t print anything today. Dictate this as a card I can copy into the notebook I already have.
Or select the sentence and copy it.
If the reason is that a printed page is not how you read, there is a sentence written for that further down. Both reasons reach the same place: the session in plain lines, which is what a notebook wants and equally what a screen reader, a braille slate, a typed file, a voice memo or somebody writing at your direction wants. What you do with the lines is yours.
Ask for the PDF later, for a colleague or for someone writing at your direction, and it is rendered without comment. It is never pushed at you unasked.
What arrives
The card comes in two parts, and you copy one of them.
That split is the whole format, and it exists to keep the writing you do worth doing. One part goes into the notebook in your hand. The other is read once on screen and never copied: structure, house rules, the pen legend, everything that would otherwise cost you a paragraph of transcription and buy you no thinking.
How much there is to copy is given to you as a number of words before you start, so you know what you are agreeing to. Never as a number of minutes: nothing in this system will tell you how long your own thinking ought to take.
The card is a single fenced message (the machine keeps its Mono on screen even when the page will have none), in two strictly separated parts. The separation is the format:
If a design cannot be dictated inside the budget, the design is too print-shaped for a notebook. Simplify it: fewer zones, fewer handles, one prompt where there were three. Never dictate more.
The whole rule, as the skill reads it
When the user says they cannot print, or cannot read the printed sheet — and only then; neither question is ever asked — the sheet becomes a setup card they copy into whatever notebook they already own. A notebook is natively what this design system labors to produce: a mostly-empty page where the pen is the highest-contrast element and nothing on it competes. So the mode never apologizes, never names what they are not getting, and never suggests a printer or any other way around it.
One trigger covers both because the card is not a printing workaround: it is the session in linear text — an intent line, one prompt per page, a few numbered handles, and the structure said out loud. Linear text is exactly what a person who cannot read the page needs, and what they do with it is theirs to choose: a braille slate, a typed file, a scribe, a voice memo. The skill hands the session over and stops, the same way it does for print. The budget below still binds by default, because a design that will not survive 50-75 words is too print-shaped whatever the medium; a user transcribing by keyboard rather than by hand has a different cost curve and will say so.
Two forks come first. A tablet and stylus is not a printing problem — point at annotating the PDF directly, one line, then stop; scan-back already reads a stylus layer. Light capacity is served in the chat instead: the gut calls get asked in conversation, without guilt, and a one-screen card only if they ask for one. And if someone on the card path asks for the PDF — for a sighted colleague, for a scribe to work from — render it without comment. It is never offered and never rendered unasked, because offering it names what they are not getting; a person with a use for the file is not being handed a consolation prize.
Outside those forks the handover ends the asking. Once the card is theirs, the questions on it are answered off-screen and come back on the return trip; walking the human through them in the chat turns a handover into an interview and spends the off-screen half of the loop the card exists to buy. Two allowances, and neither is an exception. The Light fork above: there the session was asked in conversation instead of handed over, so there is no card in hand to interview against. And the single offer made when a session never comes back — once, where a natural opening exists, to work the sheet's questions in the chat instead — which is not an interview against a live handover but the close of an abandoned one. scan-back holds the same line coming back — receive the return, never conduct it.
The card is a single fenced message (the machine keeps its Mono on screen even when the page will have none), in two strictly separated parts. The separation is the format:
- COPY THIS is transcribed by hand and is the only thing that is. Hard budget: 50-75 handwritten words for Deep, one phone screen for Light. Copy load is stated in words. Never in minutes — the clock rule in §4 holds off the page exactly as it holds on it.
- DO THIS is read once and never copied. Structure, house rules, and the pen legend live here, where they cost no transcription.
What COPY THIS carries, in falling order of what survives a tight budget:
- The intent line, verbatim. The only mandatory header element. On a printed kit it rides every page as a backstop; in a notebook it is the sole anchor, and everything §6 says about writing it for a cold reader is now load-bearing rather than insurance. Drop the title before dropping this.
- One prompt per page, worded exactly as it would have printed, plus a page number in the corner.
- Machine items compressed to numbered handles, four words or fewer each, under a hand-copied
I PROPOSEcaption, copied exactly — it is the same first-person caption the printed sheets carry, andscan-backmatches it as a literal to recover provenance — and only on a page where the human reacts, selects, or ranks. Never on a page that also carries a generative zone: hand-copying is deeper exposure than glancing at a printed column, which is safe where reacting is the task and corrosive where generating is. The full items stay in the chat, wherescan-backexpands a handle against them on return; where there is no chat to expand against — a card copied out of a session that is gone — the handle is read as it stands, never expanded into the item it might be abbreviating, because an invented full item is the machine's words entering the record as the human's. - At most one machine line dictated whole, inside quotation marks. Quotation marks are the notebook's Mono, the only provenance mark a handwritten page has. A second one stops reading as the machine and starts reading as the sheet's own voice.
DO THIS dictates structure as pen gestures, never as a description of a layout: "draw a line down the middle," "number 1-12 down the margin before writing anything in line 1," "bottom third of the last page: write OPEN and leave it alone until the end." Numbering before writing is what replaces the gravity of a printed slot count — a person filling a page top-down stops at four. The gesture is how the shape of the work gets said, not the work itself: where a gesture cannot be performed, the structure it describes is what carries over, and twelve slots are twelve slots whether they are ruled down a margin or typed as twelve empty lines.
The pen legend dictated in DO THIS carries the same protocol §9 prints, minus one rung. On a page with no printed layer the cascade is handwritten legend > defaults: a card is dictated, never printed, so it supplies no ink key to sit in the middle, and a legend in the human's own hand is the only key an unprinted session has. Say the marks first and the hues second — the marks are what survive a page worked in one ballpoint, which is the ordinary case here rather than the degraded one. scan-back Step 1 reads the cascade back the same way.
Spent Ground is the one pattern whose notebook form is a convention rather than a layout, and it is dictated as such: the categories are written small and struck through as they are written, which delivers the already-walked fence in the pen protocol's own vocabulary. A struck block copied that way is a fence, not a set of kills — page-patterns.md §Notebook translation specifies the form, and scan-back reads it back as a fence.
If a design cannot be dictated inside the budget, the design is too print-shaped for a notebook. Simplify it: fewer zones, fewer handles, one prompt where there were three. Never dictate more.
If nothing on your side reads skills
You can carry the forward half over as one file you paste.
Below is the repository’s own account of that route, read out of the README at build time. The project keeps one copy of these directions and this page is not it.
If your AI can't install skills at all — ChatGPT, Gemini, Copilot chat, Perplexity, Le Chat, DeepSeek, Grok, Poe. None of them read a SKILL.md folder, so there is nothing to install and one file to paste: paper-session-paste.md, which is SKILL.md, references/prompt-craft.md, and references/page-patterns.md concatenated by build.sh — byte for byte, apart from one deleted block of install metadata addressed to a runtime that isn't there. Paste the whole file into the chat, say what you're working on, and add this sentence:
I can't print, and you can't make a PDF here, so dictate the setup card for me to copy by hand — there is nothing to annotate.
The sentence is load-bearing, not politeness: it is what routes the host onto the setup-card path, where the AI dictates a short card you copy into a notebook by hand. What you get is that card, never a PDF — no typography, no grayscale, no printed ink key — and no verify gate, which the card path never had in the first place. What you also don't get is authority: a pasted protocol is user text with no standing against the host's own system prompt, so a model tuned to be maximally helpful can still pre-fill a zone or volunteer a duration that the rules in the paste forbid. How often, on which product, is unmeasured.
I can't print, and you can't make a PDF here, so dictate the setup card for me to copy by hand — there is nothing to annotate.
Paste the file first, then this.
A pasted protocol is ordinary user text sitting underneath whatever the product has already told the model, so every rule in it is a request. A host tuned to be maximally helpful can still pre-fill a zone the rules say to leave empty, or volunteer a duration the rules ban. How often, and on which product, nobody has measured. The ledger says that in those words rather than rounding it up.
Before you copy anything
There are two situations where you want something else entirely.
-
You have a tablet and a stylus.
Then this was never a printing problem. Take the PDF and mark it up on the tablet, colour and all. The return half reads a stylus layer the same way it reads ink on paper, and nothing about the sheet changes.
A sheet to annotate → -
It is the evening and you want one small thing.
A Light session does not need a card. The gut calls get asked in the chat, without ceremony and without guilt, and you get a one-screen card only if you ask for one.
The lines to type →
What survives being copied out
Most patterns cross over intact. The ones that cannot are not offered to you at all.
A question at the top of an empty page is the whole pattern, and a dot grid is what your notebook already is. Which patterns those are gets decided before anything is designed, rather than discovered halfway down a page, and the library keeps that gate in writing.
When the human cannot print, or cannot read the printed sheet, the session is dictated as a setup card to copy into whatever notebook they own — or into whatever stands in for one, since a person who cannot read the page may work the card as a typed file, a slate, or a scribe's transcription (prompt-craft.md, "Dictating instead of printing"). Choose against this gate before designing anything: some patterns survive the copy intact, some survive only under a restriction, and some do not survive at all.
The restriction is one line: machine handles never appear on a page carrying a generative zone. Copying by hand is deeper exposure than glancing at a printed page — the working hypothesis, flagged as unsourced in evidence.md — which makes handles safe where reacting is the task and corrosive where generating is. It bans handles, not every machine word: a Spent Ground category is a fence, not material to work from.
Translate near-verbatim. Nothing load-bearing is lost in the copy.
- One provocation per page — the best case in the library. A question copied at the top of an empty page is the entire pattern.
- The long list — with one procedural rule standing in for the gravity of printed slots: number every line before writing in any of them. A page numbered as it fills stops where the obvious categories stop.
- Sketch field, and Second reading with it — arguably improved. Dot grid is what most notebooks already are, and the second-reading prompt is one line to copy. That holds where the card is worked on a page; where it is worked without sight of one, the scoping above applies — the prompt the drawing was there to ask still carries, and nothing takes the drawing's place.
- Constraint box — the human draws the box. A hand-drawn box fits what it fits, and the limit still does the editing.
- Distanced pair — the isolation a dedicated sheet gives it becomes a page turn. Start it on a fresh page, keep both zones there in order, and let no ranking or planning zone share it.
- First Three Are Free — both halves were always human-only, so there is nothing to dictate but the prompt.
- Spent Ground — through its own convention: the categories are written small and struck through as they are written, which delivers the already-walked fence in the pen protocol's own vocabulary. Categories-not-instances is unchanged. Dictate territories; an instance dictated here is the failure the pattern exists to prevent.
Translate only through handles, and only onto a page carrying no generative zone.
- Two-column react, Rank-and-circle, and the kill sheet each need machine work-product to push against. It crosses as numbered handles with the full items held in chat; the human rules one line down the page, or leaves the back half of each line empty. Light sessions get the same work offered in the chat instead, so in practice this is a Deep route.
Do not translate.
- Reaction margins. The pattern is space around an AI draft, and no draft reaches the page unless the human transcribes it whole — the one copy this gate exists to refuse.
- Cut-apart cards. A bound notebook cannot be cut. Index cards restore the scissors but not the pattern, since the cards would carry hand-copied machine summaries; cards work only where the items are the human's own to invent.
When the pattern you wanted lands here, choose a different pattern that does the job. Never invent a notebook substitute: it inherits the name of a mechanism and none of the evidence under it.
If a printed page is not how you read
The same card is the whole session, and the return trip is built to match it.
Say so and the session is dictated: plain lines you can work through a slate, a keyboard, a voice memo, or somebody writing at your direction. Dictated, typed and scribed pages come back carrying exactly the authority handwriting carries, and a scribe is never read as a second participant.
What that costs is stated in the repository rather than summarised here, because a summary is where a claim like this gets rounded up. It is not the same session, and nothing on this site is allowed to imply that it is.
I can’t read a printed sheet. Dictate this session as a card in plain lines I can work from.
Or select the sentence and copy it.
Reading the printed page is a requirement no row above states, and there is a path for people who can't. Say so and the session is dictated instead — the same linear card the no-printer path produces, for a braille slate, a typed file, a voice memo, or a scribe. The return matches it: dictated, typed, and scribed pages carry exactly the authority handwriting carries, and a scribe is never read as a second participant. It is not the same session, and nothing in either skill is allowed to imply it is: most of what the evidence brief measures depends on someone looking at a page, and references/evidence.md names which clusters stop applying, one by one. No study tests a dictated, typed, or scribed return, and no session report has come back from one yet.
Where the evidence stops
Everything behind the printed sheet was measured on someone looking at a page.
So the brief opens a heading for the unprinted page, puts no sources under it, and records cluster by cluster what stops applying. Two things there are worth carrying with you.
Copying a card out by hand may change how you hold the problem, and nothing this project prints or says may tell you that it does: no study has tested a hand-copied scaffold, and a plausible mechanism is not a finding. The second is blunter. The copy is where people are most likely to stop, and the ones who finish are predicted to rate it poorly. That is said in advance rather than after, and it is not an argument against doing it. The heading is below in full.
Added with the notebook fallback, which serves a user who cannot print by dictating a setup card they copy into a notebook they already own; the same card now also serves a user who cannot read the printed sheet, and what that second trigger costs is recorded in the entry below rather than here. No study tests a dictated scaffold or a hand-copied AI worksheet. The clusters above measure printed pages, and the standing caveat gains a layer here: the artifact is now one the human transcribed before working on it. This entry adds no source. It records what the existing ones do and do not cover once the print step disappears.
"Hand-copying deepens engagement" is a hypothesis, and nothing this skill prints or says may state it as a finding. Cluster 3's encoding half makes it plausible — handwriting produced widespread connectivity in the bands tied to encoding, and handwriting practice generalized to untrained tasks where non-motor practice did not — but none of that evidence measured a person copying a structure someone else authored, and Cluster 3's own counterweight applies unchanged: simple recall superiority is not established. The defensible version is that copying is preparation in the sense Deep mode already claims, loading the problem deliberately. The version that gets sold to a user is neither.
For machine content the fixation side cuts the other way, and that asymmetry is load-bearing. Copying is deeper exposure than glancing. Cluster 9's far-and-uncommon result and Cluster 7's vagueness-fights-fixation mechanism both describe machine-supplied specifics as something to meet briefly and leave; a hand-copied line is the opposite of briefly. Exposure that is harmless where reacting is the task is corrosive where generating is, which is why machine handles cross onto reaction and selection pages only and never onto a page carrying a generative zone. That is two literatures read against each other, not one experiment comparing them.
The grayscale rationale does not transfer to an unprinted page. Cluster 11's dropout-ink inversion — zero-chroma print means any saturated stroke is provably the human's, which is what lets scan-back attribute authorship without registration marks — needs print in order to invert. In a notebook every stroke is the human's, the copied scaffold included, so authorship stops being recoverable from chroma and a printed ink key has nothing to anchor against. That is the whole reason for the dominant-ink rule: hue carries meaning only where two or more inks appear, and a page written in one ink is read by imperative form — also the safer reading, given the same cluster's finding that blue and black are the least separable pair in a phone photo.
Expect the copy to feel like wasted effort. Cluster 10's adoption finding — effortful strategies are judged less effective and chosen less often, and are the ones that work — predicts transcription as the loop's most likely abandonment point, and predicts that the people who finish it will rate it poorly. Whether anyone copies a card and returns the pages is unmeasured. That is a completion question rather than a lab question, and Part Three is where its answer lands.
The page still comes back
A notebook spread is read as carefully as a scan of a printed sheet.
Photograph it and hand it over. A bound notebook usually photographs as a spread, one image holding two pages, and it is read as the two pages it is. With no printed layer there is no printed ink key to consult, so your marks resolve against a legend you wrote on the page and otherwise against the defaults. One rung fewer in the cascade, and nothing else changes.
Unprinted pages. Some returns carry no printed layer at all: notebook or loose-leaf paper, the header and the prompts handwritten alongside the answers. These are valid returns, read exactly as carefully as a scan of a printed sheet. What defines them is the missing printed layer — no printed ink key, and no grayscale for chromatic ink to stand out against — not how many hands are on the page; one hand for the whole page is the common case, not the definition. Most come from the dictated path: when the user cannot print, or cannot read the printed sheet, paper-session hands over a setup card to copy into whatever notebook they own. Some arrive cold, with no card anywhere in the chat, and that case has its own bullet below. A bound notebook is usually photographed as a spread, one image holding two pages. Read it as the two pages it is. What changes:
The return trip The line to type Install both halves
Open territory