Literature Intelligence

Where It Lives

Each CATO project has a Literature area with Deep Search for retrieval and analysis and Library for durable curation. Search results, saved works, evidence outputs, fetched papers, and generated literature artifacts remain connected to the project rather than to one browser session.

Deep Search

Deep Search runs one ranked, meaning-based search over an index of scholarly publications and returns up to 100 results in relevance order. Your question is sent as you wrote it: there is no keyword translation, Boolean operators and quotes are not interpreted, and a full sentence retrieves better than a few terms.

1. State the research question as a sentence, naming population, intervention or exposure, comparator, and outcomes where relevant.
2. Optionally narrow the publication years. The range is applied during retrieval; a record the index could not date is left out.
3. Run the search and read down the ranking. Nothing is screened out for you, so review the full list rather than trusting the top few. On an enterprise organisation the search runs across the sources cleared under its agreement, every candidate is screened against the question, and only accepted works are shown, up to 100.

ClinicalTrials.gov registrations and patents are looked up through Chat, which can also search the literature directly during an analysis or writing task.

Results, Deduplication, and Enrichment

Results are enriched before they are shown or saved: missing identifiers, venue, volume and page details, and open-access links are filled in from open scholarly databases. A preprint and the journal article it became are collapsed into one result. Ranking reflects how well a record answers your question; it is not a measure of study quality or truth.

• Open a record to verify title, authors, venue, year, DOI or source ID, abstract, and links.
• Save eligible or promising works to Library; do not rely on transient result ordering.
• Record explicit reasons for exclusion outside a casual discovery workflow.

Full Text and the Workspace Reader

CATO resolves legally available open-access copies through sources such as PMC, Europe PMC, repositories, and open-access metadata services. A successfully fetched paper can open as a durable Workspace artifact for reading and supported annotation. When full text is unavailable, the interface identifies the available metadata or abstract rather than treating it as the complete paper.

Library and Collections

Library stores the works you want to keep across searches. Organize them into collections, inspect normalized identifiers and source links, add notes, and open available paper content. Tree and grid views support different curation styles.

• Use collections for protocol stages such as Candidate, Included, Excluded, Background, and Methods.
• Keep exclusion rationale separate from informal notes if it will enter a review audit trail.
• Resolve duplicate or conflicting metadata before citation export or manuscript generation.
• Prefer stable DOI, PMID, PMCID, preprint, registry, or repository IDs over title-only matching.

Evidence Matrix

Evidence extraction converts selected works into comparable structured fields such as population, intervention or exposure, comparator, outcomes, design, sample size, effect information, and potential bias or limitations. Extraction output is persisted as a research artifact so it can be reopened and used by later review or writing steps.

• Run extraction on a curated, manageable set rather than every search result.
• Open the supporting passage for every decision-critical field.
• Distinguish “not reported” from “not found in the text available to CATO.”
• Correct design and outcome mismatches before synthesis.

Citation Network

Citation relationships from available metadata can be rendered as an interactive graph with centrality, communities, and related-work exploration. Networks help identify influential clusters, bridge papers, and disconnected subfields; they do not replace critical appraisal and are biased toward sources and eras with richer citation metadata.

Watch Alerts

Save a useful search as a watch to check for new matching records. Current production allowances are:

External indexing and connector timing can delay a new record even when a watch runs on schedule.

Send Literature to Chat

1. Select works from results or Library.
2. Select Send to Chat.
3. Choose Ask with context, Analyze and extract, Draft a section, or Plan next steps.
4. Review the context packet and prefilled prompt, then remove irrelevant sources or add a focused question.
5. In Chat, open cited sources or fetched-paper artifacts before accepting the synthesis.

Context packets help bound the task. Explicitly say “use only the attached sources” when that constraint matters, and ask CATO to flag claims that lack support in the available text.

Generated Literature Artifacts and Saved Reports

Evidence tables, briefs, review drafts, citation exports, and related research outputs are stored as durable project artifacts. Saved reports render as finished documents and can be reopened at any time from the Saved reports control in the Library; a report you no longer need can be deleted from its dialog after an in-app confirmation. Citation exports support BibTeX, EndNote XML, and CSV.

Systematic Review Synthesis

The systematic-review workflow performs multi-pass synthesis of a curated evidence set, including study-design recognition, narrative synthesis, quality and limitation prompts, and structured outputs. Every claim in the generated report is traced to the works it cites. It is available on Pro, Pay-as-you-go, and Enterprise accounts.

Run Status and Failures

Searches, extractions, and report generations show progress and warnings as they run, and they continue durably if you navigate away: return to the Literature area and the finished results are delivered. Read any warnings before interpreting a small result set.

Search and Synthesis Checklist

• Save the exact question, date, year range, and any filters used.
• Check indexing dates, spelling variants, controlled vocabulary, and identifier formats.
• Resolve duplicate preprint/journal versions and retractions or corrections.
• Verify eligibility, extraction, and risk-of-bias judgments against full text.
• Confirm every generated citation and quotation in the authoritative source.
• Record where only an abstract or metadata, not full text, was available.
• Export the search and evidence artifacts required for reproducibility.