Observation, inference, and historical reconstruction
An observation records something about the world. An inference connects observations to an explanation. A historical reconstruction proposes a sequence of past events that could account for the present.
Keeping these distinctions visible helps readers ask which part of an argument is measured and which part depends on a model.
Three tiers, held apart on purpose
Direct observation. A measurement, a reading, a recorded feature of the present world. An observation can, in principle, be checked by repeating it or by an independent instrument. It does not, by itself, explain how the observed feature came to be; that step is already an inference.
Interpretive or theoretical framework. A structure of assumptions connecting observations into an explanation. A framework can be well-tested within the conditions where it was established and extrapolated well beyond them elsewhere. Extrapolation is not disqualifying on its own, but it should be named rather than absorbed silently into the word "observed."
Untested or under-tested proposal. A conjecture that has not yet faced a severe, discriminating test, or that has so far been checked only against the evidence it was built to accommodate. A proposal earns the confidence its actual testing record supports, and no more than that.
A single sentence can slide between these tiers without announcing it, for instance treating a model's output as though it carried the same warrant as the measurement that fed it. This site tries to keep the tiers separated in its own writing and treats a reader catching a place where it fails to do so as a legitimate objection, not a nuisance.
Programme structure
Several of the research programmes referenced from this site (see research programmes) organize their own claims in explicitly Lakatosian terms: a hard core of central commitments, a protective belt of auxiliary hypotheses that can be revised without abandoning the core, and severe tests meant to discriminate between competing accounts rather than merely illustrate a preferred one. These terms are defined individually in the glossary.
Naming a claim's tier, and its place within that structure, is a discipline this site tries to apply consistently. It is not a guarantee that any particular claim survives the discipline; that judgment belongs to the evidence and to the programme's own review record, not to the label.
Updating an explanation
Bayes' theorem expresses how evidence updates a probability:
Here, \(H\) is a hypothesis and \(E\) is evidence. The equation alone does not supply the priors, likelihoods, or warrant for a particular historical model. Those require explicit argument and evidence, stated openly enough that a reader can contest them. A prior chosen to secure a preferred conclusion, or a likelihood computed only under a favored hypothesis, does not survive contact with this equation; it merely wears its notation.
What would change our assessment?
Look for discriminating predictions, severe tests, alternative explanations, and open problems. A public summary should preserve the confidence and limitations of its source programme, including the parts that count against it.