The most recent 20 comments posted to Making Light by Ima Pseudonym:

Show all comments by Ima Pseudonym.

Posted on entry Talk, don't spin ::: September 02, 2007, 03:09 PM:
A number of years ago, I worked for a company that made modeling software. I'm probably still bound by the confidentiality agreement, so I'm gonna be careful here.

One day, one of the regular participants on the company mailing list / customer forum wrote that when he ran standard-textbook-case X through the model, he got result A instead of the expected result B; he wanted to know what was up? After a brief flurry of checking and testing, we discovered that a logic error in the software sometimes resulted in an error in the math. The numbers were usually not badly off, but occasionally there was a concatenation of errors which led to a result which was qualitatively wrong.

The company president felt that we should just fix this in the next software release ("all software has bugs") and keep it quiet. He thought it would look bad if word got out that our software gave incorrect results.

I felt that we had a responsibility to let all of our customers know about the problem -- they were relying on the results of the model in research papers, to decide what to do for development, etc. This wasn't a bug which would lead to a crash, or wrong colours showing up on a screen display, or something obvious; this was a modeling system which sometimes reported numbers and results that were somewhat wrong. The word was already out (on our mailing list!) and I thought it was better to admit the problem, apologize, and do what we could to rectify the problem, getting a reputation as a company that dealt honestly with a mistake, than to risk getting a reputation as a company that tried to cover things up. And this was, dammit, a standard textbook calculation; if anything, it was somewhat surprising that the problem hadn't been noticed long before.

In the end, I got my way... sort of. There was a not-very-prominent description of the problem on the company website, which got more-hidden fairly quickly. There was a brief note, not highlighted, in the newsletter sent to some customers. And of course there was the mailing list. We made a software patch available on the website (and mailed it on request), along with a script which could be used to check a given model file to find out if it would have been affected by the error. But IIRC, no effort was made to contact even the customers who'd sent us contact information on the software registration cards, let alone to seek out our customers by other means.

This attitude was one of the things that helped push me to leave the company.

A couple of years later, I discovered a couple more somewhat-wrong-numbers bugs in another of that company's software packages. When I reported the problem, I was told that they'd see about fixing it in the next release.

Comment statistics for Ima Pseudonym on the Making Light blog

YearNumber of comments posted
20071

Total: 1 comments. View all these comments on a single page.