Skip to content
System One

Span

A span is a stretch of the input text, identified by its start and end offsets. An extraction model such as Clairic answers a text field with a span, so the value is a verbatim copy of the input rather than generated text, and your code can point back to exactly where it came from.

Full guide: What is a System One extraction model? Extraction models explained

In a Clairic response, a text field returns the matched text as its value plus a span of [start, end]. In the Workers AI model page’s invoice example, the invoice number comes back with the span [8, 16]. A list field returns an array of values in input order, with their spans.

Spans matter for two reasons. The value cannot be a paraphrase or an invention, because it is a slice of what you sent. And you can highlight the source in a review screen, or check it in code, without searching the text again.

Only text and list answers carry spans. Numbers, dates and durations can be computed rather than copied. A total that adds two figures from an email, for example, appears nowhere in the email.

Decision models have no spans: they return one of the options you listed. The extraction model entry and the extraction model guide cover the rest of the answer shape.

Related terms