Enum: EvidenceSource¶

Which locator placed a set of spans in the source text.

Permissible Values¶

Value	Meaning	Description
model_quote	None	The extraction model was shown the paper and asked to copy the supporting sen...
retriever	None	The cross-encoder ranked sentence and table-row units against the value and c...
literal_match	None	The value occurs exactly once in the document and this is the clause around t...
repair_pass	None	A later pass over the built record placed this span: either resolving the val...

Slots¶

Name	Description
source	Which locator produced this set

Identifier and Mapping Information¶

Schema Source¶

from schema: https://neurostuff.org/schemas/neuroimaging-study-extraction

LinkML Source¶

name: EvidenceSource
description: Which locator placed a set of spans in the source text.
from_schema: https://neurostuff.org/schemas/neuroimaging-study-extraction
rank: 1000
permissible_values:
  model_quote:
    text: model_quote
    description: The extraction model was shown the paper and asked to copy the supporting
      sentence; `build` then resolved that quote to offsets.
  retriever:
    text: retriever
    description: The cross-encoder ranked sentence and table-row units against the
      value and cleared its margin gate, contributing a second set the model's quote
      did not already cover.
  literal_match:
    text: literal_match
    description: 'The value occurs exactly once in the document and this is the clause
      around that one occurrence, found by search before any model was asked. Distinct
      from `model_quote` because nothing read the paper to choose it, and from `repair_pass`
      because it is the first thing tried rather than a later correction. The strongest
      of the four to read: the span provably contains the value, and re-running the
      pass returns the same span rather than another sample.'
  repair_pass:
    text: repair_pass
    description: 'A later pass over the built record placed this span: either resolving
      the value''s own text against the document, or applying a resolution the extraction
      model gave with a quote. Distinct from `model_quote` because the extractor did
      not cite it, and from `retriever` because nothing ranked it -- so a reviewer
      weighing it knows it rests on a pass that came after the reading it supports.'