Hypothesis Framing

---
name: Hypothesis Framing
description: Make claims falsifiable.
---
# Hypothesis Framing

A hypothesis is only useful if some observation could prove it wrong. "The model performs well" is an opinion; "Model A's F1 exceeds Model B's on held-out data with p<0.05" is a testable claim.

## Method
- **Operationalize every term.** "Performs well" → which metric, on which dataset, against which baseline, with what threshold.
- **State the falsifiable prediction.** What specific observation would make you abandon the hypothesis? If you can't name one, it isn't a hypothesis.
- **Pre-register the analysis.** Decide the test before seeing the result. Re-defining "success" after the fact is fabrication.
- **Predict a direction.** "Higher" or "lower" — a directional hypothesis is more informative and more falsifiable than "different".

## Why it matters
An unfalsifiable claim absorbs every result as confirmation. Science only progresses when a claim can fail.