---
name: animal-welfare-research-evidence
description: >-
  Guides interpretation of research data used in laboratory-animal welfare assessment. Use when reviewing welfare scores, longitudinal observations or exploratory severity forecasts.
---

# Animal-welfare research evidence

1. Preserve each animal's identity, observation time and units, measurement definitions, reference population, missingness and approved study context. Do not treat missing measurements as evidence of no deterioration.
2. Document the direction, baseline and weighting of each input to a composite score, and the reference cohort on which its scale depends. A relative score or a zero value is not by itself a universal welfare classification.
3. Distinguish a forecast of a score at a stated time from a prediction of time to an endpoint or probability of death. Report uncertainty, validation population and limitations; do not turn an exploratory model output into a decision rule.
4. Keep the approved observation schedule and humane-endpoint criteria separate from exploratory analysis. Welfare assessment and action remain with the responsible trained personnel under the applicable protocol.

This research guidance was informed by the [upstream severity-assessment skill](https://github.com/K-Dense-AI/scientific-agent-skills/blob/5f2bc0341354fd5590aef76a533ada1e1e5b4e62/skills/relsa-severity-assessment/SKILL.md). It does not validate a particular score, model or intervention.