Criterion E6 — Novel Prediction

Validity type	External
Pass condition	The mechanism predicts new, previously untested behaviors that are then confirmed
Evidence family	Behavioral, Causal
Minimum reporting	The prediction, when it was made, the test, the outcome
Common failure mode	Post-hoc explanation of known behavior presented as prediction

What this criterion requires

Novel prediction is the strongest form of external validity. It asks whether understanding the mechanism allows the researcher to predict something that was not known before — and then confirm it. Post-hoc explanation of already-observed behavior does not count.

Satisfied when:

1. A prediction is stated before the test. The prediction follows from the mechanistic account and was not used to construct it.
2. The prediction is specific. It names a behavior, a condition, or a quantitative range.
3. The prediction is tested and confirmed. The predicted behavior is observed under the predicted conditions.

MI example

If the IOI circuit account is correct, it predicts that specific heads should activate differently on sentences with repeated vs. non-repeated names, even in syntactic structures not used for discovery. Testing this on novel constructions and confirming the predicted activation pattern would satisfy E6. Few MI studies attempt novel prediction — most work explains known behaviors rather than predicting new ones.

Connection to the chain

Required for Validated tier. Novel prediction distinguishes a mechanistic account that explains from one that understands — the difference between fitting existing data and anticipating new data. It is the criterion most directly borrowed from philosophy of science (Popper, Lakatos).