mosaic compare

Builds a table comparing cached papers along dimensions you choose (method, dataset, metric, result by default), with the LLM reading each title and abstract.

Usage

mosaic compare --query "hall thruster" -n 4 --dimensions "method,dimensionality,key finding"
mosaic compare --query "hall thruster" --sort citations -n 15 -o comparison.md
mosaic compare --from refs.bib --dimensions "method,code,validation"
mosaic compare --query "plume" -o comparison.csv

Behaviour

Papers come from the cache: those matching --query (every word in the title or abstract), those listed in a --from file (by DOI), or the whole cache. --sort citations or --sort year orders them before the first --max (default 20) are taken.

Without an LLM (llm.provider and llm.api_key set), only the metadata columns (year, source, journal, DOI) are filled and a notice says so; the LLM dimensions show –. If the LLM call fails, the same fallback applies.

--output writes .md, .csv or .json; the table is printed in any case. Exit code 1 for another extension.

See also: tutorial 8, compare, AI recipes.

Options

mosaic compare [OPTIONS]

Option	Value	Default	Description
--query, -q	TEXT		Filter papers from the cache by title/abstract
--from	PATH		Load papers from a .bib or .csv file
--max, -n	INTEGER	20	Maximum number of papers to compare
--dimensions	TEXT		Comma-separated comparison axes (default: method,dataset,metric,result)
--output, -o	PATH		Write table to file (.md, .csv, .json)
--sort	TEXT		Pre-sort papers before comparing: citations or year