Reading the laboratory
How we describe evidence
Every experimental claim the laboratory makes carries one of five labels. The label is part of the claim. This page says what each label means, and — more importantly — what it does not.
| Label | Means | Illustrative example |
|---|---|---|
| KnownKnown | Directly established facts about the laboratory, its software, its configuration, or an experiment. | The experiment ran under configuration version X for the stated interval. |
| ObservedObserved | Measurements or behaviors recorded during experiments. | Between the stated intervals, the recorded frequency of behavior Y rose from A to B. |
| InferredInferred | Interpretations supported by observations but not directly measured. | The rise in Y is consistent with the subject retaining information from the earlier interval. |
| HypothesizedHypothesized | Testable explanations or predictions that have not yet been established. | If the earlier interval is withheld, Y will not rise. (Testable; may fail.) |
| UnknownUnknown | Questions for which the available evidence does not justify a conclusion. | Whether the rise in Y reflects anything beyond the specific history that produced it. |
The examples are illustrative of the labelling scheme only. They do not describe any experiment.
What the labels do not mean
- Observed
- Observed is not synonymous with explained. A recorded behavior is a fact about the record. Why it occurred is a separate question, and usually a harder one.
- Inferred
- Inferred is not synonymous with known. An inference is the laboratory's reading of the observations. A different reading may fit them equally well.
- Hypothesized
- Hypothesized is not synonymous with expected. A hypothesis is stated so that it can fail. Stating one is not a prediction that it will succeed.
- Unknown
- Unknown is an acceptable scientific result. When the evidence does not justify a conclusion, the honest label is Unknown, and the laboratory uses it.
Practical rule. A claim may be promoted — from Hypothesized to Observed, or from Inferred to Known — only by evidence, never by rewording. When a claim's label changes in a released document, the change is logged as a correction. See Research Integrity and Research Disclosure & Reproducibility.