Glossary

A quick reference to the vocabulary used across these docs. Terms are grouped by where they appear: the core primitives, the evidence and trust language, the literature-landscape map, and the parts of a finished brief.

#Core terms

#Research brief

The primary output of BioSkepsis: a structured, fully cited answer to a single research question, rather than a chat reply or a list of links.

#Agent run

The autonomous process that produces a brief — the agent plans the question, searches the literature, reads full text, weighs the evidence, and writes the answer, showing its work at each step.

#Sub-question

A focused piece the agent decomposes your question into; each is worked through in turn and marked Covered, Limited evidence, or Not established.

#Curated search

The agent's pass over the broad life-science literature (40M+ papers) that weights authoritative, top-tier sources and matches on the biological concepts a paper discusses rather than keywords alone.

#Full-text reading

Reading entire papers — methods, results, and discussion — instead of abstracts alone, so the agent captures caveats, conditions, and context behind a result.

#Foundational paper

A highly connected, field-defining paper that a research theme rests on; the anchors of the literature landscape (see anchor / bridge / novel-lead paper).

#Evidence & trust terms

#Evidence type

The label attached to every finding describing the kind of evidence behind it, which tells you how far the claim can be pushed. The four types are:

• Causal — an intervention or mechanism establishes cause and effect.
• Associational — a correlation is reported, without established causation.
• Synthesis — drawn from reviews or meta-analyses across multiple studies.
• Preclinical — from in vitro or animal models rather than clinical data.

#Synthesized finding

A finding corroborated by two or more independent sources, tagged Synthesized to mark that it does not rest on a single paper.

#Trust Index

A single 0-to-100 score, with a trust band such as High trust, summarizing how much to rely on a brief based on how well its underlying evidence holds up — not a grade on the writing.

#Facet check

One of the six checks behind the Trust Index — provenance, grounding, evidence strength, reproducibility, durability, and reagent validity — each marked OK, Caution, Weak, or N/A.

#Verified citation

A numbered citation whose claim the agent checked against the source's full text, marked full-text-verified rather than abstract-only.

#Literature-landscape terms

#Biological fingerprint

The reduction of a paper to the biology it actually discusses — genes, proteins, drugs, diseases, pathways, and GO and MeSH terms — used to connect papers whose fingerprints overlap.

#Anchor / bridge / novel-lead paper

Three roles a paper can play in the landscape: an anchor is a foundational paper a theme rests on; a bridge connects one theme to another; a novel-lead paper is an emerging result at the edge of a field.

#Theme

A cluster of connected papers — found by community detection — representing a research area the field is organized around.

#Momentum

The trajectory of a theme, shown as rising or fading, so you can tell active areas from settled ones.

#Mechanistic links table

A structured table drawn from the landscape that lays out proposed mechanistic relationships — which entities act on which, and through what pathway — grounded in the cited sources.

#Output terms

#Hypothesis

A candidate research direction the agent surfaces from gaps, contradictions, or novel-lead papers in the landscape — a testable proposition, not an established finding.

#Methodology

The account of how a run was produced — the sub-questions planned, sources searched and read, and how evidence was weighed — recorded so the brief can be inspected rather than trusted blindly.

#Not reported (NR)

A marker used where a source does not state a value or detail the agent looked for, keeping the absence explicit rather than guessing or silently omitting it.