In a June 2014 cover story for Fortune, I helped raise to prominence the inventor-entrepreneur Elizabeth Holmes and her remarkable—I think everyone will still go along with that adjective—diagnostics company Theranos.

Fairly high up in my story there is a whopping false statement. After explaining that Theranos’s tests could be performed with a finger-stick, rather than using traditional venipuncture (a syringe in the crook of the arm), I wrote that the company “currently offers 200—and is ramping up to offer more than 1000—of the most commonly ordered blood diagnostic tests, all without the need of a syringe.”

Sixteen months later, John Carreyrou of the Wall Street Journal published a now famous front-page story containing a wide range of unflattering accusations about Theranos. Among them, he reported that one “former senior employee” had told him—in an account generally corroborated by three other former employees—that as of December 2014, the company was actually performing only about 15 finger-stick tests using its proprietary technology; the remainder were being performed using conventional, third-party analyzer machines, made by companies like Siemens—i.e., the same machines used by conventional labs like Quest Diagnostics DGX and LabCorp LH .

In that Journal article, a Theranos spokesperson was quoted flatly denying the newspaper’s allegations in a blanket manner, but refusing to be specific, citing trade secrets. Notably, from my perspective, she did not say how many tests the company had, in fact, been performing by proprietary methods in December 2014.

In a longer statement issued the same day on its website, Theranos further blasted the Journal for, among other things, relying on the accounts of “anonymous, disgruntled former employees,” but still declined to state how many proprietary tests it had really been performing.

It wasn’t until a week later—on Oct. 22—that Theranos, after stonewalling and threats of legal action failed to quell the furor, offered a serious, 14-page response to the Journal article, addressing the full panoply of its accusations. Among other things the company asserted that, as of December 2014, it had in fact been performing “more than 80 of the tests on our online test menu via finger-stick,” and that all but “a few” of those “ran using proprietary technologies.” It also asserted that in the fourth quarter of 2014, 57% of all tests ordered had been performed by finger-stick.

Those figures, if accurate, would suggest that the company might well be accomplishing, as it has claimed, something genuinely innovative and beneficial to society. I should add that the company’s extremely low prices and price-transparency would also be of unquestionable benefit to society, even if the company weren’t doing anything technologically innovative. On the other hand, all these advances matter only if Theranos’s tests are also reliable, which the Journal article also cast doubt upon. In my opinion, the evidence for this was weaker than its evidence that the company was misrepresenting its accomplishments.

After Theranos’s Oct. 22 statement, I contacted officials there to inquire how many tests it had actually been commercially performing by proprietary means in June 2014, when I had reported that it was offering more than 200. A spokesperson acknowledged to me that it was fewer than 200, but declined to specify how many. She said she’d send a statement.

The company’s statement arrived Nov. 3. “As discussed when you visited Theranos,” it said, “Theranos could perform hundreds of tests (more than 200) using its proprietary technologies. The reports you reviewed at Theranos covered many of those tests Theranos developed for use with finger-stick samples.”

(The company had, indeed, provided me in the Spring of 2014 with what it said were—and what looked to be—validation studies for scores of different diagnostic tests, though, in truth, I lacked the expertise to assess their significance.)

The statement then went on to address my followup question, which was basically: If you were capable of doing 200-plus tests using your proprietary methods, why weren’t you in fact doing them?

Here, with trademark, Theranos-ian opacity, is the reply:

Over time, we’ve been optimizing our clinical lab to bring up tests that are more commonly ordered, and in some cases move resources off the proprietary tests that are less commonly ordered to get to a point where the ordering patterns we are seeing can all be accommodated through our finger-stick technology.

Got that?

I then started looking back at my research for the original story—which had been conducted, by then, 17-19 months earlier—to try to reconstruct how I made the error.

As much as I’d like to say that Holmes lied to me, I don’t think she did. I do believe I was misled—intentionally—but I was also culpable, in that I failed to probe certain exasperatingly opaque answers that I repeatedly received.

“We do routine, specialty, and esoteric tests,” Holmes told me in May 2014. “What we’ve done is take those, and develop the chemistry and analytic systems that made it possible to run them on a microsystem.”

When I started my research in March 2014, there were maybe something like 100 tests listed on Theranos’s online test menu. The number was gradually climbing as my work continued. By the time I was ready to publish there were 214 tests listed. I assumed that meant they had now adapted 214 tests to run on their microsystem.

In fact, at the time I didn’t know what else it could have meant. That’s because, so far as I can tell, at the time of my research the company had never revealed that it ever used conventional, nonproprietary analyzers to perform the tests it listed on its menu other than for research purposes.

About a year later—in May 2015—Holmes did explain to me, as I reported then, that the company had begun using such third-party analyzers, analyzing blood drawn by venipuncture. It had begun doing so, she explained, in order to be able to perform less commonly ordered tests, which she referred to as “esoteric” tests, thereby providing customers with one-stop shopping. This was presented to me as a new development, however—a departure from prior practice.

In mid-September 2015—after an inspection at which an FDA official opined that the company’s tiny blood collection device, known as a nanotainer, was an “uncleared” medical device—Theranos discontinued using its proprietary methods for all but at most one test, for which it had already received full FDA approval. This fact didn’t come to light until a month later, about a day after Carreyrou’s first article, and just a few hours before his second.

Back in early 2014, on both its website and in press releases of that time, Theranos did disclose that “we can use a tiny finger-stick or collect a micro-sample from a venous draw” (emphasis added). It thereby acknowledged that it was capable of doing venipuncture, but stressed that even then it strived to take only micro-samples. In our conversations, Holmes defined the word “micro-sample” as referring to the “tiny samples” or “these tiny droplets we take.”

Thus, “micro-samples”—whether from finger-stick or venipuncture—were what the company was about, and my assumption was that in either case they were being analyzed by the same proprietary devices, which were designed specifically to read tiny samples.

The company’s ads, promotional materials, and posters hanging on the headquarter walls, all seemed to hammer home this point. They featured slogans like “One tiny drop changes everything”; “All the same tests. One tiny sample”; and “1/1000th the size of a typical blood draw; Theranos runs any test available in central laboratories.”

My interviews with Holmes in 2014 seemed to further reenforce the point. “When we do collect venipuncture,” Holmes explained to me back then, “we take a smaller sample than traditionally required and we also use the smallest needle. It’s a tiny butterfly needle. The least painful. So it is a smaller volume.”

Surely this meant they were still analyzing these tiny draws with their proprietary devices—the ones uniquely designed to do so, I thought.

Nevertheless, the truth is that there was at least one great, flapping red flag that I missed. By May of 2014, I had already heard about anecdotes in which people had shown up at one of Theranos’s Wellness Centers and had been disappointed to learn that the test they ordered would require a venipuncture draw.

So I asked why that sometimes happened.

“The biggest reason,” Holmes told me in May 2014, “is we’re scaling. As we’re building out this infrastructure, we’re also building out our inventory and our capacity in terms of the number of samples that we can handle at any given point in time. . . . We’ll use venipuncture in addition to the micro-samples just to handle the volume of sample that we’re processing.”

I couldn’t understand why venipuncture would help cope with volume. Were they running out of nanotainers or some other specialized piece of equipment needed for fingerstick draws at some of the Wellness Centers? Running short of some reagent? Or was it just that they hadn’t yet adapted certain tests for their proprietary platform?

“It’s more how we configure our own analytical systems,” she said, “and the capacity that we have in those systems at any given point in time. And it evolves and it’s changing. Every week we have more and more capacity.”

I remember coming back to this point again later: How exactly would venipuncture help them cope with volume?

I was told that my question was getting into the realm of trade secret.

A secret, for sure. But maybe a different kind of secret.

So I blew it. And I should have included all these colloquies in the original story. I regret the error.