---
title: "Prompticle Zero"
subtitle: "The format, explained in the format."
date: 2026-09-25
version: "1.0"
tags: ["meta","prompticles","writing","format"]
summary: "A prompticle is a short, source-backed article in four fixed parts, written to a published, hash-named set of instructions. This one defines version 1.0 by being one."
status: "published"
number: 0
variant: "a"
model: "claude-fable-5-1"
context: "CF50A7"
---
## Spark
Define a blog format that fits how people read now: they skim first, verify only when it matters, and hand the next question to a language-model assistant. Name the parts, explain why each exists and what it must and must not do, and demonstrate the whole thing by writing the definition in the format itself.
## Delve
A prompticle is a short article with four fixed parts, always in this order: **Spark**, **Delve**, **Spurs**, **Marginalia**. The Spark is the question, written as a prompt. The Delve is the answer, with the most useful sentence first and everything after it in descending order of importance. The Spurs are numbered prompts that carry the topic further, and any of them can become the Spark of the next prompticle. The Marginalia is the bibliography and endnotes: every source the Delve leans on, ordered by how much weight it carries, plus a record of how the piece was made. Stop here and you have the format.[^stamp]
**Why a new format.** Readers on screens scan rather than read, and eye-tracking showed the scan has a shape: attention pools on the first lines and the left edge.[^1][^2] Search taught us to expect the answer in the first result, and chat assistants in the first sentence. Newsrooms solved the same problem a century ago with the inverted pyramid; the military restated it as "bottom line up front," and consulting as the Pyramid Principle.[^3][^4][^5] A prompticle is the inverted pyramid rebuilt for a reader with an assistant open in the next tab.
**The Spark is a prompt, not a headline.** It is one to three sentences written as a brief to a colleague or a model: what to produce, for whom, under what constraints. It must be self-contained, specific enough that two competent writers would produce recognizably similar answers, and pasteable, so a reader can drop it into a model for a second opinion. A Spark may be a Spur lifted from an earlier prompticle.
**The Delve answers the Spark, then stops.** Its first paragraph stands alone. Each paragraph after it matters less than the one before, so the reader decides how deep to go. The tone is a capable assistant asked to be concise: plain, direct, no throat-clearing. Concise is not thin: a claim that carries weight gets a numbered reference to the Marginalia, and a claim that needs a paragraph of caveats gets the paragraph or gets cut. Most Delves run 300 to 900 words.
**The Spurs keep the question open.** They are four to seven numbered prompts, each pasteable into any assistant with no other context, ordered from easiest to hardest and tagged with a tier letter: **R** retrieve, **S** synthesize, **C** create, **F** frontier (questions nobody can yet settle). The numbering matters, because a Spur is cited by number when it becomes someone else's Spark. Vannevar Bush imagined readers leaving trails through documents for others to follow in 1945; a Spur is a trail head.[^6]
**Context versions make the generation reproducible.** A prompticle is usually drafted by a model, so the instructions it was given matter as much as the Spark. They are published as a context version: a document named by the first six hex characters of the SHA-256 hash of its own text, such as {CF50A7}, so it can never be quietly edited; change a word and it is a new version. Every prompticle names the instructions it was written with as the second entry in its Marginalia. A reader who wants to turn a Spur into a new prompticle can copy those instructions, or write their own, and give them to a model along with the Spur.
**The Marginalia makes the Delve checkable.** It is a bibliography and endnotes in one, ordered by how much of the Delve rests on each source, and each entry says what kind of source it is and which claim it supports. It opens with two fixed entries, the generation stamp (date, model, platform) and the instructions used, and closes with a provenance note: what the human wrote, what the model wrote, what was checked. A Delve with an empty Marginalia is an opinion piece, allowed as long as it is visible.
**Version 1.0, in one screen.** A prompticle is:
- A Markdown file with a short metadata block: title, date, version, series number and variant letter, tags, summary, model, context hash, and parent Spur if any.
- Four second-level headings, in order: Spark, Delve, Spurs, Marginalia, with nothing before the Spark.
- Spurs as a numbered list, each a bold tiered title and a fenced text block.
- Citations as Markdown footnotes, `[^n]` in the text and `[^n]:` under Marginalia, the first two being the stamp and the context.[^7]
- Alternates of one number told apart by letter (0a, 0b), never by editing a published version.
The file is the canonical artifact; this page is generated from it.
**What this version does not settle.** Front-loading suits explanation and argument and fights narrative, tutorials, and anything whose point is the journey. A prompt-shaped Spark can pull the writing toward answer-shaped prose when the honest answer is "it depends," so the Delve may say that first. The largest open question is recursion: if Spurs become Sparks and models answer them automatically, nothing in the structure stops a chain of plausible text from drifting from the truth; the Marginalia is the only anchor. Whether the format improves understanding, rather than the feeling of it, has not been tested; Spur 7 asks how it could be. The parts most likely to change are the tier letters, the metadata fields, and whether unanswered Spurs get collected for anyone to pick up; when they change, so do the version number and the context hash.[^8]
## Spurs
1. **[R] Retrieve · How reading changed**
```text
Summarize how online reading habits changed from early-2000s blogs, through search engines, to chat assistants built on language models. What does a reader now expect from the first paragraph of an article, and which studies or measurements support that expectation? Cite your sources with titles and years so I can check them.
```
2. **[R] Retrieve · Answer-first traditions**
```text
Trace the history of "answer first" writing structures: the journalistic inverted pyramid, the military "bottom line up front" (BLUF) convention, and Barbara Minto's Pyramid Principle. For each one, explain what problem it was created to solve, give a short example of it in use, and describe a situation where it is known to fail.
```
3. **[S] Synthesize · Compare dense formats**
```text
A "prompticle" is a short web article with four fixed parts: a Spark (the question, written as a prompt you could paste into a chat assistant), a Delve (the answer, most important sentence first, paragraphs in descending order of importance), Spurs (numbered follow-up prompts a reader can copy into an assistant), and Marginalia (a bibliography and endnotes that say what each source supports). Compare this format with three existing dense formats: the Twitter/X thread, the "atomic essay", and the explainer newsletter. For each, say specifically where it beats the prompticle and where it loses, then say which audience each one serves best.
```
4. **[S] Synthesize · Where front-loading hurts**
```text
Answer-first writing (the most important sentence first, everything after it in descending order of importance, as in the journalistic inverted pyramid) suits explanation and argument. List the kinds of writing it damages, such as narrative, tutorials with prerequisites, and persuasion, and explain why in each case. Then, for a four-part article format made of a question, an answer, a list of follow-up questions, and a list of sources, propose how each part should adapt for those kinds of writing without abandoning the structure.
```
5. **[C] Create · Write one**
```text
Write a prompticle on a topic you know well and can source. A prompticle has four parts in this order: Spark (the task, written as a prompt), Delve (the answer, under 600 words, most useful sentence first, paragraphs in descending order of importance), Spurs (five numbered follow-up prompts of increasing difficulty, each usable on its own with no other context), and Marginalia (numbered sources, each saying what kind of source it is and which claim it supports; cite only sources you are confident exist). Begin the Marginalia with a note giving the date, the model, and the tool you are running in.
```
6. **[C] Create · Validate one**
```text
Design a validator, as a specification and then as code in a language of your choice, for Markdown files in a "prompticle" format. A valid file has: a YAML metadata block at the top; exactly four second-level headings in this order: Spark, Delve, Spurs, Marginalia; non-empty Spark and Delve sections; a Spurs section that is a numbered list where every item starts with a bold title beginning with a tier letter in square brackets (R, S, C, or F) and is followed by a fenced code block with the language tag "text"; and a Marginalia section that contains a footnote definition (like [^1]: ...) for every footnote reference (like [^1]) used anywhere in the file. Separate the checks that can be fully automated from the editorial judgments that need a human, and write the exact error message for each automated check.
```
7. **[F] Frontier · Recursive drift**
```text
Imagine a publishing system where every article ends with numbered follow-up questions, a language model answers each question automatically as a new article, that article ends with its own follow-up questions, and the cycle repeats without human review. What, if anything, stops the chain of articles from drifting away from the truth while staying fluent and plausible? Propose a concrete mechanism that would keep such a chain grounded (for example, source requirements, verification steps, or halting rules), then design an experiment with measurable outcomes that would show whether the mechanism works or fails. Be explicit about what result would count as failure.
```
## Marginalia
[^stamp]: **Generation stamp.** Generated 2026-09-25 at 17:19 EDT (21:19 UTC) by Claude Fable 5.1 (model id `claude-fable-5-1`, Anthropic), working in Claude Code (desktop app, Code tab) on macOS. Each regeneration of this prompticle replaces this note.
[^1]: Jakob Nielsen, "How Users Read on the Web," Nielsen Norman Group, 1997. . Usability-study report, a primary source for its own findings; supports the claim that readers scan rather than read.
[^2]: Kara Pernice, "F-Shaped Pattern of Reading on the Web: Misunderstood, But Still Relevant (Even on Mobile)," Nielsen Norman Group, 2017, updating the 2006 eye-tracking study. . Eye-tracking research report; supports the claim that attention pools on the first lines and the left edge.
[^3]: "Inverted pyramid (journalism)," Wikipedia. . Reference article; supports the description of the newsroom structure the Delve borrows.
[^4]: "BLUF (communication)," Wikipedia. . Reference article; supports the claim that "bottom line up front" is an established military-writing convention.
[^5]: Barbara Minto, *The Pyramid Principle: Logic in Writing and Thinking*, first published 1987, later editions by Pearson. Commentary and method, a secondary source; supports the claim that conclusion-first structure arose independently in consulting.
[^6]: Vannevar Bush, "As We May Think," *The Atlantic*, July 1945. . Primary source, the original essay; supports the claim that reader-made trails between documents were proposed in 1945.
[^7]: GitHub Docs, "Basic writing and formatting syntax: Footnotes." . Published documentation, a standards-like source; supports the footnote syntax used here.
[^8]: **Provenance.** The site's author wrote the Spark, devised the four-part structure and its names in private notes between 2024 and 2026, and approved the instructions before this regeneration. The model wrote the Delve, the Spurs, and this Marginalia to those instructions. The URLs in notes 1 through 4, 6, and 7 are public pages the model is confident exist; the author had not re-verified them as of the stamp date. No empirical study of the format has been run.