# Claim Evidence Alignment

> Use when integrating results into a paper or drafting any paper section — enforces that every claim maps to specific evidence before it can be written

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

---


# Claim-Evidence Alignment (Discipline Layer)

## Overview

A claim without evidence is an opinion, not a result. This skill activates at Phase 5 (results integration) and remains in effect through Phase 6 (paper writing).

**Core principle:** Evidence first, claim second. Never the reverse.

**Violating the letter of this rule is violating the spirit of this rule.**

## The Iron Law

```
EVERY CLAIM MUST MAP TO SPECIFIC EVIDENCE. UNMAPPED CLAIMS ARE DELETED.
```

## The Claim-Evidence Mapping Table

Before writing any paper section, build and maintain this table:

| Claim | Evidence Source | Figure/Table | Statistical Support |
|-------|---------------|--------------|---------------------|
| (exact claim text) | (experiment, dataset, condition) | (Fig X / Table Y) | (test, p-value, CI) |

Every row must be complete. Empty cells mean the claim is not ready to write.

## When This Runs

1. **Before paper writing begins (G4 gate):** Full table must exist and pass review.
2. **After each paper section is drafted:** Re-check that no new unmapped claims crept in.

## Rules

- Every claim intended for the paper must link to a specific figure, table, experiment, or statistical test.
- Claims with insufficient evidence must be either: (a) deleted, or (b) supported by running additional experiments. No third option.
- Claims with weak evidence (e.g., p=0.08) must use weakened language — "suggests" not "demonstrates", "indicates" not "proves".
- The methodology-reviewer agent checks claim-evidence alignment independently. Passing your own check is necessary but not sufficient.

## Red Flags — STOP

- Claim not linked to any figure or table
- Statistical test not reported for a comparison claim
- Using "significantly" without a significance test
- Evidence column says "see results" instead of a specific reference
- Claim language stronger than the statistical support warrants

## Rationalization Prevention

| Excuse | Reality |
|--------|---------|
| "The trend is clear visually" | Visual trends need statistical backing. Show the numbers. |
| "Everyone knows this is true" | Common knowledge still needs citation. Your claim needs evidence. |
| "The result is obvious from the method" | Theoretical arguments need empirical validation in a paper. |
| "I'll add the evidence later" | No claim without evidence. Write evidence first, claim second. |

## The Bottom Line

```
No evidence → no claim
Weak evidence → weak language
Strong evidence → then and only then, strong claim
```

Map every claim. Delete the rest. No exceptions.