---
name: research-loom-synthesize
description: Use when combining research search branches into an evidence synthesis with traceable claims, disagreements, and remaining gaps.
---

# Evidence Synthesis

Read source ledgers and [references/evidence.md](references/evidence.md). Merge by DOI, repository identifier, or verified bibliographic identity; retain all goal/branch provenance. Link preprint and published versions rather than treating them as independent studies.

For each focused question, organize evidence by themes, methods, contexts, and disagreements. Map claims to sources and supporting locators. Separate source findings from interpretation; abstract-only evidence supports only explicit abstract-level claims. Explain comparability and uncertainty; do not count more papers as stronger evidence by itself.

For each research paper, explain the method and how the authors validated it, then assess whether that validation applies to the focused question. Read [references/evidence.md](references/evidence.md) for the small assessment record and triage categories. Ground method and validation details in inspected passages; distinguish absent validation from inaccessible evidence. Triage each gap with a reason and useful next action, without automatically discarding a paper.

Deliver a per-goal synthesis, claim-evidence map, access/quality limits, and prioritized gaps. Route terminology/coverage gaps to search planning; route a changed question to narrowing and user confirmation. Otherwise state whether evidence supports the requested deliverable and return the synthesis to the orchestrator. Finish at synthesis for a research brief; suggest outlining only when a paper is requested. Do not assert novelty or causal conclusions beyond the evidence.

When invoked by Research Loom, return paper/search/claim records matching [the shared paper data contract](../research-loom/references/paper-data.md). The orchestrator merges and validates them.