---
name: evidence-grounded-synthesis
description: Use when answering a research question by combining evidence, interpreting conflicting findings, or assessing a proposed research claim.
---

# Evidence-grounded synthesis

Connect the conclusion to the evidence that bears on it. The answer should make it possible to see what is observed, what is inferred, and what remains uncertain.

## Process

1. Frame the question and its relevant context. Identify the claim or decision that the synthesis should address, along with boundaries such as population, physical regime, setting, or date.
2. Gather and inspect suitable sources or supplied data. For external research, use an appropriate current search when required and verify consequential citations. Record limits on search coverage and source access. If the question calls for a systematic review, specify and perform the needed protocol and selection work before using that label.
3. Organize evidence by the particular claim it supports or challenges. Distinguish direct measurements, model predictions, simulations, reviews, and expert interpretations. Compare methods and conditions before combining findings, and account for overlapping datasets or studies rather than counting them as independent confirmations.
4. Weigh the support, alternatives, and uncertainty. Explain relevant differences in design, assumptions, scale, and precision; do not treat the number of papers alone as the strength of evidence. Separate absence of evidence from evidence of absence. Identify unsupported steps between a result and a broader claim.
5. Answer the question at the level warranted by the evidence, link consequential claims to accessible sources, and describe meaningful disagreement or limitations. Identify which sources or claims could not be independently checked. Identify the next observation, calculation, or source that would most improve confidence. If evidence is insufficient, say so plainly instead of supplying a definite conclusion.

Do not infer that an unsearched topic is novel, turn a correlation into a causal result without justification, or describe proposed work as a completed finding. Report which searches, computations, and checks were actually performed.

## Decisions informed by evidence

When the user needs to choose, identify the available options, the decision-maker's objectives, constraints, time horizon, and criteria. Include a relevant baseline or status quo if it is a realistic option. For each option, describe expected benefits, costs and burdens, feasibility, and possible adverse outcomes using evidence appropriate to that setting. Separate measured outcomes from projections and value judgments. Identify whose costs and benefits are counted and over what period; do not compare numbers with incompatible populations, units, or horizons as though they were equivalent.

Use an editable comparison table with columns for option, criterion, evidence and source, expected consequence or range, uncertainty and missing information, and implementation constraints. Record criteria weights only if supplied or explicitly proposed as assumptions. Where a recommendation depends on uncertain inputs or preferences, explain which changes could alter it; a sensitivity analysis is performed only if the relevant calculations or comparisons were actually done. Mark unavailable costs, effects, and probabilities as unknown rather than filling gaps with plausible figures. If evidence or preferences are insufficient, present the decision as unresolved and identify the information that would help.

## Arguments and counterarguments

For a debate or rebuttal, define the proposition and scope before organizing competing positions. Represent each substantive position in terms its proponents would recognize, including the strongest relevant support, assumptions, and counterevidence. Separate a source's findings from the conclusion drawn from them, and distinguish factual disputes from disagreements about values or definitions. Check whether an apparent contradiction reflects different settings, measures, or comparison groups. Do not manufacture symmetry when the evidence is uneven, or infer that an unaddressed argument is false.

Offer an editable argument map with columns for claim or objection, supporting evidence and source, assumptions or inferential steps, strongest relevant response, and unresolved limitations. When drafting a rebuttal, answer the particular objection directly, cite only checked evidence, acknowledge points that remain valid, and identify any proposed revision separately from a change actually made. Preserve contrary findings and clearly label missing evidence, untested counterarguments, and positions reconstructed from incomplete material.

These tables can be proposed as working templates without research access. Populate them with actual findings and decisions only when the relevant material has been supplied or examined, and distinguish a suggested analysis from one performed.