# Write Evidence Grounded Results

> Draft scientific Results, figure captions, abstracts, or discussion passages from structured evidence without overstating conclusions. Use when writing or revising a manuscript from data, analysis outputs, figures, references, or a SciPhys Evidence Bundle.

- Skill: `sciphys-ai/write-evidence-grounded-results` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add sciphys-ai/write-evidence-grounded-results`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sciphys-ai/write-evidence-grounded-results/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: sciphys-ai (https://skillmd.com/u/sciphys-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/sciphys-ai/write-evidence-grounded-results

---


# Write Evidence-Grounded Results

Write from claims to evidence. Preserve the boundary between what was observed, computed, interpreted, and hypothesized.

## Workflow

1. Inventory the supplied Evidence Bundle, figures, tables, methods, and references. List missing evidence before drafting.
2. Create a claim ledger with one row per proposed sentence: claim type, evidence references, uncertainty, caveats, and citation need.
3. Draft Results in the order that best answers the research question. Lead with measurements and effect sizes, not adjectives.
4. Reserve mechanisms, implications, and alternative explanations for Discussion unless the requested venue combines sections.
5. Use calibrated language from [references/claim-language.md](references/claim-language.md).
6. Mark unresolved placeholders explicitly. Never invent sample counts, instrument settings, statistics, citations, or numerical values.
7. Return the prose together with the claim ledger so a reviewer can trace each substantive statement.

## Output contract

- prose appropriate to the requested section and venue;
- claim ledger mapping statements to evidence node IDs;
- missing-information and citation-needed list;
- limitations and plausible alternative explanations;
- no unsupported claim presented as established fact.

## Guardrails

- Use causal language only when design and analysis support causality.
- Do not turn agreement with a reference into proof of identity.
- Do not equate absence of a detected signal with absence of the phenomenon.
- Preserve meaningful uncertainty and conflicting evidence.
- Ask for expert approval before submission-ready scientific or clinical claims.