Microsoft Remote Connectivity Analyzer. CheckSite automatically checks all pages of a website's code for a number of potential problems, and reports them. This allows for easy website testing that is easily repeatable.
The people that started CheckSite have been in banking for 25 years, and in data processing since the early 1960's. Over that period of time the concept of 'garbage in - garbage out' has manifested itself time and time again.When Check 21 was first proposed the ideas was that this would be an inter-bank clearing mechanism manned by experts who knew all the ins and outs of check processing and the need for 'proof of deposit.'
The advent of general public use, even a selected general public, meant that there would be a proclivity to make errors. The depositing of checks is important to a company, but it is not something that people make a career of. Everyone knew there were going to be errors.
Check Site For Malware
When we came out with our Version 1 software in January 2005, we naturally tested the program at a bank. Each day we ran 2,000 checks through the system - not processing them, just checking the system for both errors and integration. We used employees who'd never worked with checks before, tellers who been working with checks for years, high school students completely untrained, and employees who had been completely trained to run the CheckSite system.Checks were inserted into the scanners upside down, backwards, there were 'piggy-backs' with checks on top of each other, extraneous information was also put through the scanners. In other words, if it could happen, it did, and trained people did as poorly as untrained.
With the release of Version 1.1.2, we'd made many changes. Most importantly, we gave up trying to use expensive scanners with magnetic read heads and CAR/LAR software, and went to trained experts. This switch from machine processing to 'proof of deposit' operators (actual people verifying the checks) was the key. Read the checks in any way you want - backwards, forwards, upside down - it makes no difference. Put through a recipe for Duck soup and it makes no difference, whatever is scanned is scanned, and with the verification process in place, we anticipated many fewer problems in our next test.
Our new process went like this: first, the user transmits the information. They could send one check or 300, it didn't matter. Once received by the computer at the central location, the computer would send back a confirmation number. also, within seconds of receiving the deposit, the system sends an email acknowledgement that the deposit was received and gives the depositor a deposit number.
The transmitted email acknowledgement would not show how much was deposited, or how many checks were received, only that the computer received a deposit on that days' date, and that the deposit contained a certain number of items, with the deposit number.
Next the images were parceled out to a large number of key operators, our 'proof of deposit' experts. Though certain parts of the individual checks were hidden from the operators, the amount was visible and that would be keyed as shown. Two operators got identical information, but neither knew of the other. If both keyers agreed on the amount, the check was accepted, and if there was disagreement the information was sent to a third keyer. If all three disagreed, the check goes to a fourth person, a supervisor, and the check is closely inspected.
Isthatsitedown
After the data is keyed, a second email was sent describing the user's deposit in detail. Included in this document was a table of statistics showing the results of the keying operation. In a random test of 250,000 checks, the first set of keyers keyed the full 250,000 checks and the second keyed 248,816 because some of the checks were rejected for all sorts of writer errors. From this double keying, about 1.5% went to a third keyer (3,650 checks), and of these 19 went to a supervisor. The supervisor resolved the vast majority of these, and the final error rate was under 1 basis point.
Check Site Ip
By June 0f 2005, our first venture into the real world of checks began. In the lab we were getting less than one error in 125,000 checks. But once we began using the process in the real world, we were getting much higher error rates. So we went back and reworked the system. The errors were simple once you found them. For logisitcal reasons, we were letting the first keyer and the second keyer be the same person - this was corrected immediately. We also ran into a problem understanding the difference between the number 7 and the number 7 with a line through it - a more common occurence than we thought possible.
In the late summer of 2005 we started an extensive training program, and by the time we released our Version 1.2.0 the real world error rate was reflecting the expected lab rate.