The definition has a shorter memory than the number

Somewhere in the next board pack there is a row with two years of the same metric in it. Same label, same company, one number bigger than the other. The gap between them is the thing everyone will discuss.

Almost nobody will check whether the two numbers were calculated the same way.

What specific disclosure standards does the SEC mandate for software renewals?

This is not an invented standard. In January 2020 the SEC published guidance on the metrics companies put in their management discussion, and it is specific about what happens when one of them changes:

If a company changes the method by which it calculates or presents the metric from one period to another or otherwise, the company should consider the need to disclose, to the extent material: (1) the differences in the way the metric is calculated or presented compared to prior periods, (2) the reasons for such changes, (3) the effects of any such change on the amounts or other information being disclosed and on amounts or other information previously reported, and (4) such other differences in methodology and results that would reasonably be expected to be relevant.

And separately, that a company should consider whether it needs to recast prior metrics so the current disclosure sits in context.

Note the hedging, because it matters for everything below: should consider the need to disclose, to the extent material. This is guidance about what a filer should think about, not a rule with a tripwire. Nothing here is an accusation that anyone broke one.

What nobody had done is check what the filings actually contain. That is the same discipline as preferring evidence to anecdote: the claim is only worth what the primary document says.

What do public 10-K filings actually disclose regarding enterprise customer retention?

I read four years of annual reports from a fixed group of software companies, pulled the passage where each one describes how it calculates its retention metric, and compared each year against the one before it. Then, because a similarity score is not a verdict, I read every difference by hand against both filings.

Then I did something that cost me my headline. I gave the passages to a second coder who could not see the company names, the years, or any of my conclusions, mixed in fifteen pairs that were word-for-word identical as controls, and asked it to code them cold. It caught all fifteen controls. It also disagreed with me on a third of the cases, which is the honest reason the number below is six rather than the nine I started with.

This corpus is the same one behind an earlier piece on what happened to a benchmark cohort, read for a different question.

Six definitional changes survive both readings. Not many, across 48 companies and four years. The finding is not that this happens constantly. It is what it looks like when it does.

Which SaaS issuer established an auditable standard for net retention reporting?

HubSpot changed its calculation in 2025 to remove the effect of certain partner commissions. It said so in the filing. It gave the reason. And it published what the metric would have been under the old method for both prior years: 102.2% for 2024, 103.9% for 2023.

That is the whole of the guidance, done. A reader can see the change, understand why, and compare the new number to a like-for-like history. It takes one sentence and two restated figures.

Worth being precise about one thing, because my second coder was: what the filing shows is a statement that the calculation was adjusted, rather than a before-and-after formula. The restatement is the strongest evidence in this corpus that a company did what the guidance describes, and it is still a company telling you rather than showing you.

How do selective disclosures conceal underlying customer renewal decay?

MongoDB’s definition of its expansion metric contained, for four consecutive years, a sentence about how self-serve customers’ revenue was folded into the calculation for direct sales customers. In the fifth year that sentence is gone.

I looked for a disclosure of the change across the whole filing, not just the passage: twelve sentences in that document carry methodology-change language, and the only one that mentions the metric is a risk factor about how quickly customers consume the product.

The rate moved from about 118% to about 121% across those two filings. I want to be careful here: those two facts sit next to each other and I am not claiming one caused the other. I cannot see inside the calculation, only the sentences describing it, and a definition can change for reasons that have nothing to do with the number.

What unexpected disclosure patterns emerged from longitudinal SEC filing reviews?

Two companies disclosed a change, exactly as the guidance describes, and then removed the disclosure from the following year’s filing while the change itself stayed in force.

Kaltura did it thoroughly. Its filing says the prior year’s rate “has been recast to reflect the update to our customer count methodology”, and quantifies the effect: an adjustment of one percentage point. That is the effect stated and the prior period restated, which is the whole of what the guidance asks. Confluent disclosed its change and noted the prior amounts had not been retroactively adjusted, which is less but is still telling you. In both cases the sentence is present in one annual report and absent from the next, while the changed method continues.

Nothing improper is happening, and in Kaltura’s case the disclosure that later vanished was a model of the form. A company describes a change in the year it makes it, and the following year the change is simply how things are done, so the sentence goes. It is the most natural editorial decision in the world.

But look at what it does to a reader. If you compare this year’s filing to last year’s, you see the notice. If you compare the two most recent filings a year later, you see two identical definitions and no indication that anything ever moved. The change outlives the disclosure of it, and the reader most likely to be misled is the diligent one, comparing the most recent documents.

Which emerging SaaS pricing models escape existing regulatory disclosure rules?

Box’s calculation sentence is word-for-word identical across the transition I looked at. What disappeared was a separate claim: that the metric is presented on a constant currency basis.

In the earlier filing, fifteen sentences mention the metric and exactly one says it is constant-currency. In each of the four filings after it, thirteen sentences mention the metric and none of them does.

So did Box stop presenting on a constant currency basis, or keep doing it and stop saying so? The filing cannot tell you, and I am not going to guess. What is certain is narrower and still useful: a reader could establish the currency basis of that number in one year and cannot in the next. Currency treatment is one of the things that differs between companies reporting the same metric, so losing the label is not nothing.

What diagnostic audit must investors run on reported customer retention metrics?

The instinct when reading something like this is to conclude that the numbers are untrustworthy. That is too strong and not very useful. Six changes across 48 companies over four years is a low rate, and one of the six is a company doing the disclosure properly.

Two of the six changes here were disclosed properly, one of them with the prior year recast and the effect quantified to a percentage point. It is the same failure mode as a corpus that quietly goes out of date: nothing is wrong on the page, and the page is no longer telling you what you think. The useful conclusion is narrower and it is about your own reading habit. A metric compared against its own history is only meaningful if the definition held still, and the filing will not flag it for you when it did not, because the flag has a shorter life than the change.

So when you put two years of the same number side by side, whether it is a peer’s or your own: open both definitions and read them. Not the number, the sentence underneath it. It takes a minute per comparison and it is the only way to see any of this. On the evidence here, comparing any two consecutive years of one company, it will matter roughly one time in twenty-five. Over a four-year stretch, roughly one company in eight will have changed something.

Where are the legal and empirical boundaries of software disclosure analysis?

Boundary. The corpus shows that definitions can outlive their notice, not that every KPI change is undisclosed. Re-read definitions in both filings and expand the panel before estimating a general rate.

Evidence base. The analytical frame also draws on these additional sources: U.S. Securities and Exchange Commission 2020. The links identify the exact works; they support the mechanisms and boundary conditions discussed here, not every claim in isolation.

References

1. U.S. Securities and Exchange Commission. (2020). Commission guidance on management's discussion and analysis of financial condition and results of operations (Release No. 33-10751). https://www.sec.gov/rules/interp/2020/33-10751.pdf