Extracting values

A candidate is a proposal with its evidence

Extraction reads an approved document and proposes candidate values. Each carries the value and unit as printed, the property, the protein, and the reported conditions such as treatment, heating temperature and time, pH and protein concentration.

Each candidate also records where it came from: the page and the table, sentence or figure. A candidate with no evidence is not accepted.

A candidate is not yet data

A candidate never becomes a dataset row by itself. Unit conversion, quality checks and a reviewer’s approval all come after it.

You can filter candidates by status and by whether they need review, and work through similar ones together.

What kind of number it is

The platform records what kind of number each value is, because the same digits can mean different things.

• A measured value, which is the usual case.
• A change relative to a control, which cannot be combined with measured values without the control.
• A threshold, such as "gelation occurred above 12 per cent", which needs its "above" or "below" to make sense.
• A range, which has two limits and is never reduced to its midpoint.

Repeats and new runs

The same measurement found in a table and in the abstract is shown as one value seen twice, not as two findings.

Extraction can be run again for one paper or for all eligible papers. A new run never replaces an approved value; it produces new candidates to review.