Monday, July 11, 2011

7+3 = 11

tl;dr – a boring trivial bug is causing me to procrastinate by writing about it
tool – a spreadsheet to help you choose input data to spot the particular pathology

I love moo cards*.

Here's a snap of a recent bill from moo. Spot the bug.
100 cards £9.17 ... Shipping £2.50 ... VAT £2.33 ... Total £14.01 ... new blog post: priceless


The problem is a known pathology**. It's not uncommon to find that basket calculations are sometimes off by a penny; the calculations are done with precision, and those precise numbers are fiddled to fit with our quantum of currency - the penny. The error fits the fiddle.

In this case, the total including delivery and VAT looks as if it should be precisely £14.004. Expecting this to be £14.00, one might be tempted to speculate that the total has been rounded up my mistake, but two things make me not so sure.

Ⴀ) I generally see problems related to truncations (which always go down; £14.004 -> £14.00) and normal rounding (£14.005-> £14.01, but £14.004-> £14.00).
Ⴁ) thinking about it, I had a 10% discount on the normal price as a sop for the knock on effects of a previous bug. Discounts add another layer of complexity.

Let's work the numbers:
  • £10.19 is the normal price.
  • After the 10% discount, that would be £9.171, not £9.17.
  • Add £2.50 delivery to arrive at £11.671.
  • 20% VAT on is £2.3342.
  • The precise total is £14.0052 - which will be rounded up to £14.01.
  • The VAT component is £2.3342 - which will be rounded down to £2.33.
That seems more plausible.

Is this a rare combination of numbers? I built a spreadsheet to explore, and it is not; 300 prices between 1p and £10 show this behaviour.

All this is in the context of a 10% discount, 20% tax and £2.50 delivery. But my spreadsheet is a model, so I can change the conditions. Playing with it gives me the following empirical understandings:
  • you don't see this problem without a discount;
  • within reasonable ranges, picking alternative discounts doesn't change the incidence much;
  • within reasonable ranges, changing the tax doesn't change the incidence much - I've seen it go down to 200;
  • the range of incidence seems to be 200-300 for 'reasonable' ranges of tax and discount
  • the delivery charge doesn't matter if it's to 2dp (and my model is inaccurate with 3dp)
Constraining myself to a basket with one item, I expect that I can sit down and demonstrate mathematically to my own satisfaction that in order to see a total that rounds up (ie £14.005), and an associated tax that rounds down(ie £2.3342), you need a price with a third decimal place - ie a normal price that has already been adjusted in some way. But that efficiency, while attractive, is a procrastination too far. For now, I'm happy with the general rule of thumb; you only see this problem when at least one thing in your basket can have a price that includes fractions of a penny - but if the potential is there, you'll see if for 20-30% of your possible prices.

Coders: One solution is do all calculations off-screen to full precision, but produce the totals on the bill from the numbers that actually go on the bill. Another is to round your total to 2dp before calculating tax. Of course this can mean having two containers for very similar information.

So far, so fun. For testers.

Frankly, I don't mind paying the extra penny. My problem is what the penny does to my paperwork.

I'm doing my VAT accounts, where I separate the £2.33 from the rest of the total. Moo's fluff on their bill means that stuff that should add up to zero, doesn't. I'll have to fudge the penny, which means introducing a special case. I'll have to be careful, because special cases are where I make accounting mistakes. That's a pain. I hope that you (or Moo) can use that description to advocate a fix for similar bugs.

And I hope that you go out there to find them. Here's the link to that spreadsheet again. I'll use it to generate data to help me reveal this issue***. You may use it and abuse as you wish. Please attribute me if you use it in public. It's got a second page that shows incidence, and a third with instructions, license and known bugs.




*  Those of you who have had a business card from me are charmed by them, too. Moo's custom postcards will lend excellent grooviness to a game I have in mind. I want to make special stickers for a bunch of post-it related activities. Moo have always responded swiftly and sweetly to problems, and  to top it all, they're local.
** I know this pathology, and I look for it when I test. Indeed, I've got an exercise based on something very similar in one of my classes. Some people question the veracity of that exercise; surely no-one really has obvious errors like this any more. Ha.
*** I tried a google docs version, but it runs like a three legged dog on Safari and Firefox. I was so discouraged I didn't bother with Chrome...

No comments:

Post a Comment