One Bad Image

Damian came up with a great pseudo-paradox (conundrum) the other day.  Imagine you're paper scanning one thousand documents for a client.  You complete the project, and mail off the CD's and docs, and receive payment.

One week later, the client calls and says, "I found a bad image on the CD!  The text is cut off!"

You apologize profusely.

The client adds, "This is no good.  I'm sending the whole project back.  I want it redone right this time.  IF THERE'S ONE BAD IMAGE, THEN THE WHOLE THING IS PROBABLY WRONG!"

You apologize again, and begin looking at your back-ups (just in case the client's CDs break or don't work -- it happens).  You spend two hours scrolling through all 1,000 images and ... you only find the one image that is wrong.

By Sigma-Six standards, one image out of 1,000 is a perfectly acceptable error rate.  It's .1%  To be frank, it's to be expected.

The client, who either didn't read up on Six-Sigma or doesn't care, insists that the entire job be redone.  He did not look through every image, like you did, but is convinced there have to be more mistakes.

It's like this: the client is convinced that, since he spotted a bad image without even trying to look, this statistically proves there are more bad images, even though a true statistician would say the opposite is true and request only the one page be rescanned.

Quite a pseudo-paradox, huh?

Now, imagine the same scenario, but with millions of images on microfilm or microfiche, and you'll have entered the world of Microfilm Scanning.  It takes scanning and QC to keep error rates low, and guys like Damian and I happen to be experts in that field.