# Keep uncertainty intact when editing a draft

An editor should improve the expression of uncertainty without removing uncertainty that belongs to the evidence. Watch for changes from may to will, estimated to exact, and early observation to proven result. Those are changes in meaning, even when the sentence reads more smoothly.

## Mark the strength of each claim
Label a statement as observed, estimated, inferred, or proposed. This is a working editorial aid, not a universal scientific classification. Attach the source or assumption that explains the label. A model often sees a confident sentence as better prose; your brief should make accuracy the constraint.

## A fictional comparison
| Draft | Misleading edit | Better edit |
| --- | --- | --- |
| Two testers missed the field | Users cannot find the field | Two of five internal testers missed the field |
| Estimated delivery in June | Delivery is in June | Delivery is currently estimated for June |
| Could reduce handoffs | Reduces handoffs | May reduce handoffs; not yet measured |

The better edit adds useful scope or preserves the original qualification. It does not bury every sentence under “perhaps” and “possibly.” Use the qualification that identifies the actual limit.

## Audit with this prompt
```text
Compare this original and edited draft.
Find changes in certainty, causality, scope, timing, and measurement.
For each change, quote both versions and explain the difference.
Restore uncertainty that the supplied evidence requires.
Do not add uncertainty to straightforward verified facts just to sound cautious.
Original: [paste]
Edit: [paste]
Evidence and assumptions: [paste]
```

## Keep the important qualification nearby
A disclaimer at the end of a report does not fix an unsupported headline. Put the relevant limitation beside the claim: the sample size, timeframe, assumption, or missing measurement. If the limitation changes the decision, it belongs in the summary too.

Before publishing, look specifically at titles, captions, chart labels, and pull quotes. They often lose qualifications when shortened. A reader seeing only those elements should understand the same level of certainty as someone who reads the full article. Preserve the distinction between what happened, what you think it means, and what you want to test next.

---
SkillStall · 2026-10-02