Why does the method require the hypothesis to be written before the test is run? This note explains the order, using a momentum rule on the Russell 2000 as the public example of a claim that is easy to rewrite after the fact.
The test is not the place the idea is born
A hypothesis is a sentence that can be made false. It names the rule, the population, the window, and the observation that would kill it. A test is the attempt to produce that observation honestly. If the sentence is written after the test, the sentence will be whatever survived, and the kill condition will be edited to spare it. The culture of writing it first is not superstition about paperwork. It is the only way a later reader can see the difference between a prediction and a caption.
The sentence can be short. Brevity is a virtue, because a paragraph that contains every escape hatch cannot be killed. What matters is the timestamp on the sentence relative to the timestamp on the result. A file that stores both, in order, is a method. A slide that shows the result and then the rationale is a pitch. Pitches have their places. They are not this process.
A hypothesis is a sentence, written before the test, that names the rule and the observation that would kill it. A sentence written afterward is a caption.
A ranking that wants to be rewritten
A momentum rule on the Russell 2000 is a public shape that invites rewriting. Before looking, the hypothesis might say: using point-in-time membership, a trailing window of a stated length, and a rank formed on a stated day, the rule will be evaluated on the next period, and it is killed if the apparent relationship disappears once a second, pre-specified regime is included. After looking, the temptation is to mention only the window that worked, to explain away the regime, and to call the explanation the hypothesis. The writing-first rule exists so that temptation has a document to contradict.
Nothing here says the momentum rule is true or false. The example is about the order of sentences. A researcher who writes the kill condition and then hits it should be able to point at the earlier sentence and stop. A culture that congratulates the subsequent edit, the one that finds a neighboring window, has taught everyone to write hypotheses last. Interns learn this faster than anyone, because they have not yet learned which edits the room will applaud. The applauded edit is the one to distrust.
There is a data version of the same sin. Changing the discard rule after seeing which discards improve the picture is a new hypothesis, not a cleanup. The Russell study that drops names until the rank looks better has started a search. The search has to be counted, or a later overfitting check has nothing honest to count. This note’s demand is prior and simpler. Write the sentence. Then touch the data. If you must touch the data to discover what you meant, you are still exploring, and exploration should not be labeled as a test.
Keep the earlier sentence visible
The earlier sentence should remain in the file when the result is attached. Deleting it because it looks naive next to a cleverer later thought destroys the evidence of order. Naive and prior is more informative than polished and retroactive. Reviewers should ask to see the prior sentence first, before they look at the chart. A chart seen first will recruit them.
Write the hypothesis before the test so the test can say no to something definite. The Russell ranking is the example of a claim that will try to become definite only after it sees the answer. The method is the refusal to let it.
This piece does not claim that momentum on the Russell 2000 is a hypothesis the firm holds, or that writing a sentence guarantees an honest test. It says the order is the part a culture can actually enforce. A polished caption added after the chart is not a late-arriving hypothesis. It is a different document, and it should be labeled as one so the order stays visible.
