Document 1
Document 2
| # | A | B | Type | Score | Description |
|---|
An OpenRouter account is required to access LLM models.
Don't have an account? Sign up at openrouter.ai
Choose the LLM to use for generating hilinks.
The initial atomization round — restates the document as flat sentences. Placeholder: {{docText}}
Runs once per atomizr level beyond 1, summarizing the previous level and mapping each summary to it. Placeholders: {{sentences}}, {{maxIndex}}
Partitions a document into verbatim body (kept) and heading (stripped) spans. Strips internally-facing scaffold (titles, section headings, page numbers, TOC); keeps all body prose, however short. Placeholder: {{docText}}
Cleans a reading-order section into a verbatim delete + reorder body (never reworded) plus references-with-context. Sections are cleaned concurrently and aligned back to source coordinates. Placeholder: {{rawText}}
Placeholders: {{docCount}}, {{docSections}}
Placeholders: {{docCount}}, {{docSections}}, {{seedQuestion}}
Edits the template for the current document count (2 docs vs 3+). Placeholders: {{docCount}}, {{docSections}}
Sent once per shared point that exactly two documents address, to rate their positions on a signed −1..1 scale (+1 same position, 0 neutral/orthogonal, −1 opposed; disagreement stays in the union). Placeholders: {{k}} (doc count), {{kMinus1}}, {{labels}} (document labels, e.g. A, B), {{label}} (the point/question), {{materials}} (per-document material).
Sent once per shared point that three or more documents address. Returns the most-similar and least-similar pair scores on the same signed −1..1 scale. Same placeholders as above.