# hypothesis-generation

> Structured hypothesis formulation from experimental observations or data — formulating testable hypotheses with predictions, proposing mechanisms, and designing experiments to test them following the scientific method framework.

**Use case**: Formulate testable hypotheses from experimental observations

**Canonical URL**: https://agentcookbooks.com/skills/hypothesis-generation/

**Topics**: claude-code, skills, science, scientific-writing

**Trigger phrases**: "generate hypotheses from this data", "formulate a testable hypothesis", "what mechanisms could explain", "design an experiment to test", "propose a hypothesis for"

**Source**: [K-Dense AI](https://github.com/K-Dense-AI/scientific-agent-skills/tree/main/scientific-skills/hypothesis-generation)

**License**: MIT

---

## What it does

`hypothesis-generation` is a Claude Code skill from K-Dense AI's [scientific-agent-skills repo](https://github.com/K-Dense-AI/scientific-agent-skills). It turns Claude into a hypothesis formulator that takes experimental observations or preliminary data and applies the scientific method framework to produce: clearly stated testable hypotheses with explicit predictions, proposed mechanistic explanations, and experimental designs for testing each hypothesis.

A session produces structured hypothesis cards: null hypothesis, alternative hypothesis, predicted outcome if true, predicted outcome if false, and a proposed experimental design with key controls.

## When to use it

Reach for it when:

- You have preliminary or pilot data showing an unexpected result and need to articulate what mechanisms could explain it
- You're writing the hypothesis statement for a grant application and need to make the logic from observation to prediction explicit
- You're designing a follow-up experiment and want to enumerate alternative hypotheses before committing to one

When *not* to reach for it:

- Open-ended ideation without specific observations — use `scientific-brainstorming`
- Automated LLM-driven hypothesis testing on datasets — use `hypogenic` (a separate, more automated skill)

## Install

Copy the `SKILL.md` from K-Dense AI's [hypothesis-generation folder](https://github.com/K-Dense-AI/scientific-agent-skills/tree/main/scientific-skills/hypothesis-generation) into `.claude/skills/hypothesis-generation/` in your project.

Trigger phrases: "generate hypotheses from this data", "formulate a testable hypothesis", "what mechanisms could explain", "design an experiment to test".

## What a session looks like

A typical session has three phases:

1. **Observation intake.** Describe the experimental observations — what you expected vs. what you saw, with any relevant context about the system and prior literature. Claude asks clarifying questions about controls and conditions.
2. **Hypothesis enumeration.** Claude generates 3–5 candidate hypotheses ordered from most to least parsimonious, each with a proposed mechanism and explicit testable predictions.
3. **Experimental design.** For each top hypothesis, Claude proposes a minimal experiment to test it — key manipulations, controls, and the result that would support vs. refute the hypothesis.

## Receipts

**Where it works well:**
- Generating alternative hypotheses you hadn't considered — Claude's breadth of biological and physical mechanisms surfaces plausible alternatives that a specialist locked into their own model can miss
- Making hypothesis logic explicit for grant writing — the null/alternative/prediction format directly maps to what grant reviewers expect

**Where it backfires:**
- Highly system-specific mechanistic hypotheses (e.g., specific protein interaction networks) may be superficially plausible but technically implausible in your specific model system — domain expert review is essential
- The skill generates hypotheses, not priority rankings; deciding which hypothesis is most worth testing requires your experimental judgment

**Pattern that works:** generate 5 candidate hypotheses, then apply `scientific-critical-thinking` to each proposed experimental design to identify the design flaws before committing to an experiment.

## Source and attribution

Originally authored by [K-Dense Inc.](https://github.com/K-Dense-AI). The canonical SKILL.md lives in the [`hypothesis-generation` folder](https://github.com/K-Dense-AI/scientific-agent-skills/tree/main/scientific-skills/hypothesis-generation) of their public scientific-agent-skills repository.

License: MIT. Install, adapt, and redistribute with attribution preserved.

This page documents the skill from a practitioner's perspective. For the formal spec and any updates, defer to the source repo.