Full-Text Retrieval in Research Agents: Coverage Beyond the Abstract
Abstract-only retrieval bounds the evidence available to an agent. Full-text retrieval extends coverage to methods, results, and cited context.
Most research tools stop at the abstract because the abstract is what the index serves. But an abstract is a marketing document: it reports the headline effect, not the sample size that qualifies it, the exclusion criteria that narrow it, or the confidence interval that spans zero.
A citation you haven't read is a rumor with a DOI.
What reading the primary source changes
◆ abstract-only retrieval
│ └─● headline effect, as the authors chose to state it
│
◆ full-text retrieval
│
├─● methods was the trial pre-registered
├─● instruments did the replication use the same instrument
├─● tables does table 3 contradict the conclusion
├─● supplements the exclusions that narrow the claim
└─● registrations the ClinicalTrials.gov record behind the paper
When the agent parses the full text, it can answer the questions reviewers actually ask: was the trial pre-registered, did the replication use the same instrument, does table 3 contradict the conclusion. Those answers live in the body, in the supplements, in ClinicalTrials.gov registrations — never in the abstract.