Scientific Literature Evidence
Build the smallest source corpus that can answer the scientific question, then expose what that corpus does and does not establish.
Define the evidence question
- Identify whether the task is discovery, state-of-the-field synthesis, historical or priority attribution, version comparison, source-conflict resolution, or an absence claim.
- Record the scientific scope, relevant time and version boundary, source types, and the evidence needed to stop. These modes do not share one universal preference for either breadth or recency.
- Keep a single already-supplied citation-support judgment in
$scientific-claim-audit; use this skill when constructing or comparing a source corpus is part of the work.
Build and evaluate the corpus
- Use bibliographic records to establish identity and publication state, and the paper, supplement, data, or code containing the result to establish substantive support.
- Prefer controlling primary sources. Use reviews and secondary sources for routing, terminology, and synthesis, not as substitutes when the original evidence is available.
- For every consequential source, record its role, the exact proposition it supports, its scope, and a locator when available. Separate the source's statement from your inference.
- Read source evaluation when versions differ, sources conflict, priority matters, or the result is a bounded negative search.
Synthesize for scientific judgment
- Organize the result by the scientific distinction that changes interpretation, not by search order or an unstructured paper list.
- Expose disagreements through differences in definitions, assumptions, methods, samples, regimes, or versions before reconciling them.
- State remaining gaps and search boundaries. Repeated agreement can raise confidence but does not enlarge what the underlying evidence measures.
- Return an evidence corpus and synthesis. Compose
$scientific-claim-auditwhen that evidence must determine the validity boundary of a manuscript claim, and$structure-documentationwhen durable literature-review prose is requested.