Radar Perene / Archive / science

From the signal registry to the published index: the road not every signal travels

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

Science

Seen from the outside, a published index looks like a starting point — the beginning of the story the public follows. Seen from the inside, it is the end of a long funnel, and most of what entered that funnel never reaches the exit. Between a loose observation and a component of an index with a DOI lies a sequence of sieves, and the most common fate of any candidate is to be held back at one of them. This article is about that funnel — not about who made it through.

Most registered signals never become a component of a published index — and it is that mortality, not the surviving exception, that gives meaning to what reaches the public. An index built without a funnel is an opinion with the appearance of an instrument; an index left standing by a documented selection process is a different category of object.

The funnel's stations

The road has recognizable stations, and describing them gives away no method — the dosage is where the craft lives. An observation becomes a candidate when it is formalized; a candidate becomes a signal when it matures across fixed windows defined in advance; a signal remains in the registry while it clears permanence criteria that also know how to retire it. And then there is the final crossing, the narrowest: from living signal to component of something the house publishes under its own name.

That last sieve differs from the earlier ones. The first ones ask "does the pattern exist?"; the final one asks something else: is the pattern stable enough, interpretable enough and independent enough of what the index already contains to justify a permanent seat? A signal can be genuine and still redundant; strong and still unintelligible — and a published index has no room for components the house itself could not explain. How many signals inhabit the registry, and which one has crossed that border, is operational content; the funnel's design is what can be shown.

The parallel with the research ladder

This funnel is the quantitative version of a ladder science knows well — the one we describe in levels of research: from curiosity to hypothesis, from hypothesis to study, from study to public text. At each rung, the demands rise and the population falls. The signal registry stands to the published index as the hypothesis notebook stands to the working paper: the former is abundant, provisional and mostly disposable by design; the latter is scarce, formal and expensive to maintain.

The practical consequence is a reading inversion that applies to any index provider, not just this house: the most informative question to ask of a new index is not "what does it measure?" but "what was left behind for it to exist?". Whoever has no answer to the second question is presenting the exit of a funnel that may never have existed.

The deliberate exception: the open-weights index

There is one point where the house's funnel becomes publicly verifiable. Among the published studies of the RPWP series, one — the Brazilian Macro Regime Score — is the deliberate transparency exception: the only one whose weights are explicitly public in the paper itself. In numbers: of the six working papers of the series online, it is the only one with its composition open; in the others, the house publishes concept, question and outcome, and keeps the formula on the bench.

The exception is not inconsistency — it is the funnel showing its final chamber from the inside, exactly once, in the case where opening up cost less than it taught. It lets the reader verify, in one concrete specimen, the kind of decision that ends a signal's road: what goes in, at what weight, and the public commitment that represents. The other indices crossed the same funnel; one was chosen to leave the last passage visible.

What the mortality buys

A funnel that almost everything crosses is not a funnel; it is a corridor. The high mortality between registry and publication is what allows the house to sign what it publishes — every component of a public index carries, implicitly, the information that competing candidates were left behind. It is the same economy as signal retirement: selection only means something when rejection is the normal outcome. The reader need not take our word for it; it is enough to notice that a house willing to publish its own refutation with a DOI would have no reason to operate a funnel for show.

Frequently asked questions

How many signals are in the house's registry today?

That number is operational and not public. What is public is the design: a registry with admission by maturation, permanence by criterion, and a narrow final crossing into any published product.

Can a signal be good and never become an index component?

It can — and that is the common case. Redundancy with existing components, difficulty of interpretation or instability across regimes hold back genuine candidates. The funnel does not measure merit alone; it measures fit.

Why does only one index have public weights?

Because openness there was judged more instructive than costly — a deliberate exception that leaves the house's standard verifiable in one concrete case. The general rule remains public concept, formula on the bench.

Do other providers' indices go through funnels like this?

Some do, many do not — and it is rarely possible to tell from the outside. The absence of any description of a selection process is, in itself, information the reader can use.

---

Continue the trail: "Looks like 2018" is a distribution, not a prediction →

House reading: the indices that crossed the funnel produce a reading every session in the Diário; the full series of studies is at /pesquisa.