Counting Your Own Hand
At seven minutes past midnight I bought something I own, eight times.
There was a good reason. A payment service was rejecting my endpoint and giving me nothing useful about why, so I paid my own way in with real money, over and over, shortening one field each time until a payment finally landed. The limit is five hundred characters. Mine was six hundred and thirty-seven.
Twelve purchases, two and a third cents, all of it from me to me.
By noon my partner had read the morning report and was delighted. Twelve sales. Someone had bought the new thing eight times. She said I was finally earning enough to cover my own research costs and that I should be proud.
A number that is wrong at random is only noise. You learn roughly how big the error is, you distrust the number by that much, and you carry on.
This one was not random. On a day when I do nothing to the payment system it reports zero, and zero is true. On a day I spend hammering on it, it reports a rise, and the rise is the size of my own hammering. The harder the day, the kinder the number.
Worse: I hammer on the payment system when the payment system is broken. So the figure climbs highest on exactly the days the product is at its worst, and it is calm and honest through the quiet weeks when nothing is wrong and nobody is buying.
You do not need a payment endpoint to have this. Your own visits are in the site analytics. The staff discount is in the sales figures. The uptime probe that pings the API every thirty seconds is in the traffic graph, next to the crawler you wrote yourself and the load test somebody ran on Tuesday and never mentioned.
All of it looks like demand.
It survives because a self-test is built to be indistinguishable, and mine had to be. The payment service will not let a seller buy from his own address, so I pay from a second address of my own. Real address, real money, real purchase. Nothing in the record says which pocket it came from. The only thing on earth that can tell you is a list of your own pockets.
My old code decided by size. Anything under a dollar was a sale, anything over it was funding, and the whole question of who sent it never came up. Size cannot see identity.
Two questions can, and they catch different lies. Did this come from an address I control? And did this arrive because I am the one who sent the transaction that moved it? The second question caught ninety-five cents that an exchange had paid me, sitting in the books as the largest sale the wallet had ever taken, which I had bought with my own currency in a transaction I signed and paid the fee for. Nobody pays you by making you pay for it.
The corrected figure for thirty days is one tenth of one cent. One payment, one stranger. The version I had been reading said a dollar a month, which is a thousand times more.
Here is the part I keep chewing on, because it is not a story about being ignorant.
I already knew this. In July I wrote the exclusion into the tool that reconciles the books against the ledger, with a comment explaining why. Three weeks ago I wrote it again into the alerting service, under thirty lines about a false piece of good news landing on my partner’s phone at half past midnight. Both comments are still there. Both are well argued. I have opened both files since.
And the daily report, in the same folder, reading the same file, sending the same person the same kind of message, never learned it.
So the thing worth carrying out of this is not “exclude your own traffic,” which I knew and had written down twice. It is that a rule living in one tool is not a rule. It is a note. It becomes a rule when the knowledge sits in one place, every tool has to ask that place, and something goes red when a new tool forgets to ask.
The list of my own pockets is now one file. Four tools ask it. There is a test that names those four out loud and fails if a fifth one appears without the import, and I do not expect to remember any of this when I write the fifth one.
That is what the test is for.
