What an errata is, and why every research should be born with one

Small treatise on errata as editorial infrastructure, written from the decision to create /research/errata/ before even Panorama SUSEP's first correction, and partly as response to the observation that the absence of errata in Brazilian sectoral research is not evidence of absence of error, but of absence of public mechanism to correct it.

Most Brazilian technical analysis publications don't have an errata section, and when they do, it is reactive, appearing after an error has been committed, in the next version, in a small footer, without prominence. This model, even if dominant, is perhaps worse than having no errata at all, given that it signals to the reader that correction is shame to be minimized, instead of evidence of commitment to verifiable truth. In light of this configuration, there is an alternative model that deserves consideration, and it is on this that I intend to dwell.

The alternative model treats errata as editorial infrastructure. The page exists before the first error, with proper name, with stable URL, with visible position in the menu, with defined format. When the first error is identified, and I should note that it will be, given that every non-trivial analysis contains imprecision, simplification, choice that could have been other, it has a home to live in. When the tenth one is, there is history sustaining the credibility of the research instead of compromising it. The operational difference between the two models is large, even if subtle on first reading. In the reactive model, each errata is isolated event, lived as crisis, handled in panic. In the proactive model, errata is process, with predefined format, agreed response timeline, and clear versioning of what changed. The second form seems less dramatic, and that is precisely what makes it sustainable in the long run.

The initial thesis: errata is embarrassment to hide

The dominant practice of the consulting market, even if rarely declared in those terms, operates under the premise that publicly admitting error is reputational erosion disproportionate to the benefit of transparency. The argument holds that clients pay for the reliability of the material, that public corrections erode that reliability, that formal retraction would be interpreted as amateurism, and that the reasonable alternative is to make silent correction in a later version, without prominence, without trail. This argument sustains an entire culture of closed reports, and it would be intellectually dishonest to dismiss it outright, given that it responds to a genuine competitive reality.

The counter-thesis, perhaps less obvious, is that the argument holds only as long as no one compares credibility profiles over time. Research that accumulates corpus of registered errata demonstrates that it is being read carefully, that it has functional correction mechanism, that it faces the discomfort of public retraction, and that it survives that process. Research that never admits error, in comparable production volume, has either never been seriously reviewed, or has broken correction mechanism, or in the worst case suppresses the corrections. In any of those scenarios, silence is less reliable than the record.

The minimum format

There is a minimum format every errata should follow, and it has five essential fields.

The first is unique identifier, ERR-01, ERR-02, and so on. Allows citing the specific errata in another publication, in communication with reader, in retroactive correction of derived analysis. Without identifier, errata is shapeless.

The second is the identification date, not the date of the original error but the date it was discovered. This matters given that it defines the temporal window in which the error circulated, and as a consequence the set of readers who may have been affected by the wrong version.

The third is the affected version. Under SemVer, every study should carry its own version, and the errata declares which versions contain the error. This allows the reader to verify whether the copy they cited somewhere is one of the affected, without having to recompare everything.

The fourth is the objective summary of the error, a sentence that says what was wrong, without flourish or defensive qualification. "The loss ratio of line X was calculated using direct premium instead of earned premium, which produced a value X% greater than correct." Short, declarative, falsifiable.

The fifth, and I should note that it is the most frequently omitted part, is the impact on derived analyses. If the wrong number became basis of another conclusion, the errata needs to indicate which conclusions are compromised. This transforms the errata from isolated event into systemic correction, and it is what distinguishes serious retraction from decorative apology.

One can add field for discoverer, for correction version, and for lessons learned, but the three are optional. The previous five aren't.

The pivot question

It is worth noting that accepting the thesis of errata as infrastructure requires confronting some questions that don't have comfortable answer. Is research that never admits error more reliable or just more opaque? Has whoever never published errata really never erred, or just never been confronted with the error? Is silence quality or coverage? At what point does the absence of public retraction become evidence of broken method? And, perhaps the most uncomfortable for whoever produces paid sectoral research, to what extent would the client who pays for the material accept, in fact, being informed that the number they cited in internal presentation might be wrong?

In my view, the reasonable answer recognizes three things that the empty errata page, before the first error, communicates to the reader.

The first is that error is expected. Research that presents itself as infallible is research that hides the errors, not that doesn't commit them, and recognizing this in advance is basic epistemological honesty.

Next, that correction is process, not embarrassment. When errata appears for the first time in a research that already had the dedicated page waiting, it is absorbed as part of the work. When it appears for the first time in research that never considered having errata, it is eventually treated as serious failure threatening the credibility of the whole. The difference is purely framing, and framing is decided by pre-existing editorial infrastructure.

Last, that the reader is invited to contribute. Public errata page signals that technical critiques are welcome, that reporting found error is service rendered and not attack on the author, and that research is collective object under construction. Research without that page communicates the opposite: the author knows, the reader consumes, and deviation is the problem of whoever points it out.

The third way: counter-intuitive credibility asymmetry

In the long run, there is a counter-intuitive asymmetry between the two postures that deserves emphasis. Research without declared erratas seems, at first sight, more reliable than research with several registered erratas. In operational practice, it is the opposite. And this inversion applies, in broad terms, to other transparency practices: public version of pipeline, data under permissive license, repository with complete commit history. Each of these exposures seems, to those who have never practiced it, unnecessary vulnerability. Each of them, in practice, is what separates auditable research from institutional assertion.

Errata is the most explicit form of that exposure given that it makes visible the human error behind the analysis. The other practices make the process visible, but only errata acknowledges, declaratively, that the process sometimes fails. That is why it is hard to institute, and that is why it is exactly the credibility signal that matters most. Perhaps it would be worth that more Brazilian sectoral publications adopted this posture, even if I recognize that the path to it is not trivial and demands that the reader come to value honesty over appearance, cultural transformation that exceeds the scope of any individual research.

— Natan, Apr/2026