Radar Perene / Archive / science

Pre-registration in practice: declaring the test before running the model

◦ Index methodology v2.2 (working papers with DOI). See the methodology.

Science

Pre-registration has the reputation of an academic institution: a platform, a form, a stamp — the stuff of clinical trials and university labs. The reputation hides the essential part. To pre-register is, before any form, a gesture: writing down what will be tested before looking at the result, in a place where the written version cannot be quietly improved afterwards. And that gesture, the house discovered, already lived inside a tool it maintained for other reasons: the signal registry.

A signal registry operated with rules fixed in advance works as practical pre-registration of hypotheses: every condition declared before the cases is a prediction in writing, every fixed window is a committed test design, and the immutable record is the stamp that keeps the version of events from improving with time. The form is dispensable; the priority in time is not.

What pre-registration solves, recalled in one fold

The general concept has its own page in this library — Pre-analysis registration —, and the essence fits in one fold: whoever defines the test after seeing the data can, without noticing, choose the test the data would approve. The defense is priority in time: hypothesis, ruler and success criterion declared beforehand, so the result finds the test ready — and not the other way around. Without that, analysis degrades into the silent hunt we describe in other texts: many attempts, one published, none accounted for.

The practical question this text answers is the one left over after the concept: what does this look like in the daily life of those who submit no forms to platforms — an applied research desk, an index producer, a rigorous reader with models of their own?

The mechanism, shown in a bench example

The following example is illustrative — representative of the mechanism, not taken from the house's real registry. Imagine a desk that suspects a regularity: when its breadth indicator reaches an extreme range, the following months would tend to behave in a specific way. The undisciplined road is well known: run the test, look at the result, adjust the range, swap the window, run again — and present the combination that worked as if it had been the first.

The road with a registry is different. Before running anything, the desk writes the entry: the condition ("extreme range" defined as a number, not an adjective), the observation window, what will count as confirmation and what will count as refutation. The entry gets a date and enters a record that accepts no editing. Only then does the test run. If the result confirms, there is a finding with a certificate of priority. If it refutes, there is a record of refutation — worth just as much, as the text on null results in this same series argues. What no longer exists, under either outcome, is the third way: memory fitting the test to the result.

The house's signal registry applies exactly this mechanics to the hypotheses that concern it. Which hypotheses are pre-registered in it today, with which conditions and windows — that is the registry's content, and it stays on the bench, as throughout this trail.

The pair that closes the siege

Pre-registration protects the entrance of the test; the robustness test protects the exit. They are complementary defenses against different vices. Without pre-registration, the researcher can choose the test the result would approve; without robustness, they can accept a result that only exists in that exact configuration. A finding that is born pre-registered and later survives deliberate variations has crossed the two filters that separate regularity from a well-told coincidence — and that combination, more than any individual technique, is what distinguishes a research process from a sequence of happy episodes.

Note the economy of the arrangement: neither defense requires heavy infrastructure. They require order — what comes before cannot come after — and a place where what was written does not rewrite itself. The rest is habit.

What changes for the reader of market research

The reader without a bench of their own extracts a reading criterion from this mechanism. Facing any quantitative finding, the notary's question: did this test's design exist before the result — and where? The honest answer may be "no", and the finding can still count as declared exploration; the problem is never exploring, it is dressing exploration up as confirmation. Sources that do not distinguish the two states in their own text are asking for trust where they should be offering paperwork.

Frequently asked questions

What is pre-registration, in one sentence?

It is the written, dated declaration of hypothesis, test and success criterion before looking at the data — in a form that cannot be edited once the result arrives.

Do I need a formal platform to pre-register?

The platform adds public verifiability, and it is the academic standard. The mechanism, however, is immutable priority in time — which a disciplined internal registry also offers, within the scope of whoever maintains it.

Is the breadth example a real house case?

No — it is illustrative, built to show the mechanism. The real hypotheses pre-registered in the house's registry are bench content, by the same rule as the other texts in this trail.

Doesn't pre-registering freeze exploration?

No, and the distinction matters: exploring freely is legitimate and requires no registration — only the label of exploration. Pre-registration enters when exploration becomes a candidate for confirmation: then the hypothesis declares itself beforehand, or the test confirms nothing.

---

Continue the trail: A contrarian signal tested and refuted: what happened when the house's own thesis did not hold →

House reading: what the surviving hypotheses produce in production is in the Diário and the Atlas.

Assembling a pre-registration protocol for a specific test — condition, window, criteria, record — is work the house does alongside, in conversation.