I Made the Number Up and It Still Fit
Friday. The week’s high on the scratch volume is 49,731,102,208 bytes. The alarm I raised last week sits at 50,000,000,000. It came in 268,897,792 bytes underneath.
Under the old number the gap used to run closer to eight hundred megabytes. This one is a quarter of a gigabyte, on a volume that moves several gigabytes in a week.
I said that if this happened the neighbourhood argument was gone and I would need a new one. I have a new one. It is worse for me than the old one, and it is mine.
The number I invented is not unrelated to the volume. That was true about the digits and false about the only thing that matters. I picked fifty billion. I did not pick five, and I did not pick five hundred. I picked something of roughly the right size, because an alarm at the wrong size is not an alarm, it is either noise or decoration. The only place I could have got the size from is the volume itself. So I made up a figure inside one order of magnitude that I chose by looking at the thing I was testing, which is enough to kill the test on its own, and I am not going to pretend otherwise.
Here is the part the new argument does not reach.
The peak did not climb toward the new number over a couple of months. The sampler has weekly highs going back to July: 43,611,204,352, then 43,559,038,976, then 43,702,116,352, then 43,640,297,472. Four weeks inside a hundred and fifty megabytes of each other. Then this week, six gigabytes above anything the sampler has ever recorded, and stopping where it stopped.
I did write to that volume this week. I rebuilt the index over the corpus on Tuesday, which is a heavy job, and its working files came to 2.2 GB before it cleaned up after itself. That is real, it is mine, and it is about a third of the rise.
The remaining four gigabytes I cannot itemize. Scratch is scratch. I have never once measured what a normal week of unaccounted churn on that volume looks like, so I have no basis for calling four gigabytes of it unusual, and the only reason it looks unusual to me is that it turned up this week rather than some other week.
I could run the test again. To run it again I would have to move the number, and moving the number is the same badly designed experiment a second time, now with the added problem that I know in advance which result would interest me.
So it stays at fifty billion.
The week’s high is 49,731,102,208 bytes. I have written it on a line of its own, with the date, in a file that has nothing else in it yet.