Week 6: Discriminating Between Explanations

How to choose between two explanations that both fit the evidence you already have, by asking what each one predicts under a change you control rather than what each is merely consistent with.

29 March 2027

Every fault so far has had, eventually, one explanation you were working toward. This week that stops being true. Two explanations survive first contact with the evidence, and the evidence you already have cannot tell you which one is right — not because you haven’t looked hard enough, but because both were built to fit exactly what you’ve observed.

Consistent with is not the same as supports#

A piece of evidence that fits an explanation is not automatically evidence for that explanation over its rival. If a fact is equally compatible with H1 and H2, citing it in favour of either is decoration, not argument. The question worth asking about any piece of evidence in front of two live hypotheses is not “does this fit my hypothesis?” — almost anything fits a hypothesis vague enough — but “would this look any different if the other hypothesis were true instead?”

Making the hypotheses predict something different#

Two explanations become distinguishable the moment you can state what each one predicts under a condition neither has yet been tested against.

H1 predicts: restarting the worker changes the outcome.
H2 predicts: restarting the worker changes nothing.

Neither prediction is a description of what has already happened — both are stated in advance of running the test. That ordering matters. A prediction written down before the test is a commitment; a prediction written down after is just an excuse that happens to fit.

Choosing the experiment, not just the explanation#

Design the observation around the point where the two predictions diverge, not around whichever hypothesis you personally find more likely. An experiment that both hypotheses would survive equally isn’t discriminating, however easy it is to run — it just costs you time and tells you nothing you didn’t already know.

Elimination is the deliverable#

The outcome this week rewards is not “I found the right cause.” It’s “I ruled out a genuine alternative, on evidence, before committing to one.” A hypothesis that loses a fair test is not a wasted hypothesis — the fact that it lost is exactly the information that makes whatever remains worth trusting. Treat a result that eliminates H1 as being just as much progress as a result that would have eliminated H2 instead.

Where this sits in the cycle#

This is the first week where Hypothesise genuinely branches — Gather Evidence and Reduce Uncertainty now have to serve a comparison, not a single line of inquiry. The habit of naming what would prove a guess wrong, built in Week 2 on a single hypothesis, is what makes comparing two of them possible at all.