Turn a paper into data
Configure your MASEMiner extraction
r Pearson, or odds ratios, smd standardised mean differences…). The right of the colon is the label shown in the prompt.var1/var2). Short code + a definition that helps the model map paper-specific operationalisations. Add a third :: chunk for synonyms.1: item text) so the model matches reordered/paraphrased items. Leave empty for copyrighted instruments.What do you want to extract?
You get one row per this in your dataset.
Describe what to extract
Full control — define every field, its type, and its review tab.
1. What is the unit of interest you want to extract?
The thing you get one row per in your dataset. Pick the closest template, or Other to start from scratch.
Each field has a type — most are Values; use List for an array of values, and Table only when a single field is itself a sub-table (e.g. a regression's coefficients across regressors, or a correlation matrix). "Many per paper" already gives you one row per unit — you don't need a Table for that.
Fields extracted once per paper, not per-unit. title / doi / year / authors are always included. Add additional fields below, one per line, in name: description format.
Anything that doesn't fit a structured field — domain-specific rules, edge cases, definitions, references to existing extraction protocols.
Write or paste your prompt. It should instruct the model to return one JSON object with a records array and an evidence array.
This is the exact prompt sent to the model — edit it if you like.
Choose a provider, select a model, and enter your API key, or point at a local / self-hosted model server. Your key is sent to our server only to run this extraction: it is held in memory for at most 30 minutes, then discarded. It is never saved to disk or to your account.
The test passes your key through our server to the selected provider once and does not keep it.
Extraction runs through your selected model — expect roughly ~1 minute per paper (longer for large PDFs). You can add several at once; up to ten run in parallel, the rest wait their turn.