Full Idea
Whatever could do the job of justifying an inference from the observed to the unobserved must itself be an inference from the observed to the unobserved.
Gist of Idea
Anything justifying inferences from observed to unobserved must itself do that
Source
Alexander Bird (Philosophy of Science [1998], Ch.5)
Book Reference
Bird,Alexander: 'Philosophy of Science' [UCL Press 2000], p.172
A Reaction
We must first accept that the unobserved might not be like the observed, no matter how much regularity we have, so it can't possibly be a logical 'inference'. Essences generate regularities, but non-essences may not.