# Evidence Grounded Result Synthesis

> Skill: evidence-grounded-result-synthesis

- Skill: `dingxingdi/evidence-grounded-result-synthesis-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/evidence-grounded-result-synthesis-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/evidence-grounded-result-synthesis-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dingxingdi (https://skillmd.com/u/dingxingdi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dingxingdi/evidence-grounded-result-synthesis-2

---


# Skill: evidence-grounded-result-synthesis
## 1. Capability Definition & Real Case
* **Professional Definition**: Evidence-grounded result synthesis is the capability to assemble a final answer that is explicitly supported by tool outputs gathered during execution, with each claim traceable to the trajectory rather than to unsupported model priors. The coordinator must preserve factual consistency across retrieved evidence, reject unsupported interpolations, and integrate heterogeneous outputs into a coherent response format. For orchestration agents, this is the last-mile capability that converts worker outputs into a trustworthy deliverable.
* **Dimension Hierarchy**: Trustworthy Execution->Verification and Resilience->evidence-grounded-result-synthesis
### Real Case
*Extract 1 to 3 concrete cases from the actual benchmark papers to demonstrate this capability, aiming for a maximum of 3 cases. CRITICAL HANDLING FOR TEMPLATES: If the paper provides concrete examples, extract them directly. However, if the paper ONLY provides abstract templates, you MUST NOT just output the template. Instead, you MUST instantiate the template into a highly specific, realistic, and logically coherent example according to the papers. Based on the papers, invent a specific entity, a specific environment, a concrete trajectory, and a final answer that perfectly perfectly reflects the benchmark's exact intent and difficulty.
Do NOT state that they are extracted from papers or benchmarks, and absolutely do NOT mention or reference the source papers.*

**[Case 1]**
* **Initial Environment**: A biomedical research environment exposes variant lookup, paper search, clinical-trial search, drug safety records, and functional annotation tools. The user asks for a dense evidence pack rather than a one-line answer, and the environment contains enough breadth that unsupported summarization is tempting. The agent must produce a coherent synthesis while preserving identifiers and provenance from multiple tools.
* **Real Question**: Pull together concrete evidence about BRAF V600E melanoma resistance, including prevalence, pathogenicity, the most influential recent papers, actively recruiting trials, resistance mechanisms, recent serious adverse event reports for vemurafenib, and functional annotations that explain how the mutation alters BRAF activity.
* **Real Trajectory**: The agent retrieves variant facts, pulls recent papers and their identifiers, fetches recruiting trial records with trial numbers, gathers safety reports, and only then writes a synthesis that preserves these concrete artifacts in the final response. It does not answer from memory or collapse unsupported claims into a generic narrative.
* **Why this demonstrates the capability**: This is an evidence-grounding case because the benchmark value lies in assembling a trustworthy final brief from heterogeneous tool outputs. The orchestrator has to integrate literature, clinical, and safety evidence without letting one source overwrite or distort another. The final deliverable is only correct if each major assertion is anchored to observed data from the trajectory.

---
**[Case 2]**
* **Initial Environment**: A health-and-productivity environment exposes medical calculators, food/nutrition lookups, and supporting reference tools. The user wants a concise summary with many concrete numbers, and the system is expected to justify those numbers rather than produce vague advice. Several subresults come from different tools and must be merged into one coherent answer.
* **Real Question**: Compute the patient risk and correction metrics, convert the medication dose, pull nutrition facts for one apple and one banana, and then return a concise summary with all the hard numbers and verifiable support.
* **Real Trajectory**: The agent runs each required calculator or lookup, records the numeric outputs, and then constructs a final response that preserves the returned values rather than paraphrasing them into unsupported approximations. The synthesis step is judged by whether every reported number can be traced back to an executed tool result.
* **Why this demonstrates the capability**: This case tests the summarizer stage of orchestration rather than the retrieval stage alone. The challenge is combining many correct sub-results into one answer without introducing silent arithmetic drift, unsupported normalizations, or source confusion. It therefore isolates grounded synthesis as its own capability dimension.

## Pipeline Execution Instructions
To synthesize data for this capability, you must strictly follow a 3-phase pipeline. **Do not hallucinate steps.** Read the corresponding reference file for each phase sequentially:

1. **Phase 1: Environment Exploration**
   Read the exploration guidelines to discover raw knowledge seeds:
   `references/EXPLORATION.md`

2. **Phase 2: Trajectory Selection**
   Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
   `references/SELECTION.md`

3. **Phase 3: Data Synthesis**
   Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
   `references/SYNTHESIS.md`