incremental-reveal

Incremental reveal

A human reads top-down. The paragraph resolving a confusion is not in their head yet when they hit the confusion. You read the whole text at once, so that confusion never happens to you — you resolve every ambiguity from context the reader does not have yet.

That blindness is specific. You will not notice a heading that misleads until the section under it is read, a term used before it is defined, a sentence that contradicts the previous one until a later one reconciles them, or the fact that the actual argument does not arrive until 80% of the way in. Reading the text again more carefully does not help, because the knowledge causing the blindness is already in your head.

The fix is structural, not effortful: read prefixes, in parallel, with separate readers who each know nothing.

The method

For a piece of N paragraphs, run N independent readings. Reading i sees paragraphs 1..i and nothing else. Reading 3 gets paragraphs 1, 2 and 3 — as a fresh read, not a continuation of reading 2.

1. Split. One file per paragraph in reading order, numbered 01…NN. The title is unit 01. Keep a list with the sentence introducing it. For slides or carousels, one image per slide. Strip anything that leaks the ending: frontmatter, editing notes, changelogs, the source URL.
2. Build prefixes. prefix-NN contains paragraphs 1..NN concatenated.
3. Spawn N readers at once, in a single message so they run concurrently. Each gets exactly one prefix. Cap it at 20 at a time — beyond that most harnesses queue anyway, and the marginal reader tells you little. For a piece longer than 20 paragraphs, either run the prefixes in batches or split on sections rather than paragraphs.
4. Synthesise from what the readers reported.

The reader prompt

Give each reader this, and nothing else:

Read the file <prefix path>. It is a piece of writing that may be complete, or may stop partway through — you have no way of knowing which, and you must not try to find out. Read nothing else: no other file, no web search, no looking for the source, the author, or any context. Do not spawn subagents.

You are an honest, slightly impatient reader — the kind who abandons things. Having read up to where the text stops, answer:

1. What do you think this piece is about? One sentence.
2. Anything confusing, or that made you stop and re-read? Quote it.
3. Would you keep reading, or drop off here? Why? Be blunt.
4. What question do you now expect to be answered?

Write your four answers to <out path> and reply with just that path.