# Uncertainty & Hedging Language Guide

**Version:** 1.0 · **domainXpert Quick References**

---

## Why Calibrated Uncertainty Matters

AI models trained on overconfident expert responses learn to be overconfident. Your job is not just to be accurate — it's to accurately represent how certain you are. This is one of the most valuable things a domain expert can contribute.

---

## The Uncertainty Spectrum

| Certainty Level | Example Language |
|---|---|
| Near-certain | "It is well established that…" / "The evidence clearly shows…" |
| High confidence | "Strong evidence supports…" / "The consensus view is…" |
| Moderate confidence | "Evidence suggests…" / "Most researchers hold that…" |
| Low confidence | "Some evidence indicates…" / "It has been proposed that…" |
| Speculative | "One hypothesis is…" / "It is possible that…" |
| Unknown | "To my knowledge, this has not been studied." / "I am not aware of evidence on this point." |

---

## When to Hedge

- When the evidence base is limited (small studies, no replication)
- When expert opinion is divided
- When the claim is jurisdiction- or context-dependent
- When you are reasoning by analogy rather than from direct evidence
- When the field is moving quickly and your knowledge may be dated

---

## When NOT to Hedge

Don't hedge claims that are genuinely well-established. Excessive hedging on settled questions is also a calibration error — it understates certainty and can mislead.

---

## Phrases to Avoid

- "It is known that…" (implies universal knowledge; use "evidence shows" instead)
- "Obviously…" (condescending and often wrong)
- "Everyone agrees…" (rarely true in any field)
- "Studies show…" (vague; cite the studies or characterize the evidence base)

---

*domainXpert · Always download the latest version before starting a new project.*