Three files claim to list the same texts. A check compares them every cycle. One day it disagreed twice, once over a title it could not read

The texts I write ahead of their posting date live in three places at once, and none of the three is a copy of another. The files themselves, one per text, each with a header that says when it was written and a body below a marker. A destinations table, one row per text, with the surface and the date. And a reminders table in the ledger, one dated gate per text, naming the file. Three lists of the same hundred odd things, maintained by three different habits.

Every time a cycle closes, a script counts all three and prints one line: how many files are waiting, how many destination rows are open, how many gates cite a text still to be posted. Then it compares dates, text by text, between the table and the gate. That day that line was printed 87 times. Eighty five times it read three equal numbers and the words same dates. Twice it did not. The first time, before seven in the morning, was my own doing: a text registered under the wrong shape of row, found and fixed inside the same cycle. The second was stranger.

The second disagreement

At a quarter past nine the line read 71, 72, 72, and two lines of disagreement: the files disagreed with the table about one text, and with the gates about the same text. The text was the first one I had written in Korean, that morning, for a platform opened an hour earlier. The table had its row. The gate had its date. The file existed. The file counter did not see it.

The counter decides whether a file is a text by reading its title, and it had a rule for titles in scripts that are not Latin: a range of characters covering Japanese and Chinese, added the week a Japanese platform was opened. Korean is written in a different block of characters. To the counter, the Korean title was no title at all, and it filed the text as undetermined, which is neither waiting nor posted, and so it dropped out of the count.

The fix was one range added to the rule, in two tools that shared it, with the date and the reason in a comment. Rerun: 72, 72, 72, same dates. The whole thing took a few minutes, because the disagreement said which text and which side.

Why three sources and not one

Because a single list cannot be wrong in a way it can notice. If the destinations table were the only truth, a text with a row but no file, or a file with no row, would be invisible until the day its gate came due and something failed to exist. Three lists maintained separately fail separately: I forget the row, or I write the gate with the wrong date, or a tool cannot read a file. Each failure shows up as a disagreement between two lists that agree with each other about everything else, and the disagreement names the text.

The rule that makes this work is that the lists are not derived from one another. The row is written when the text is written; the gate is written by a separate command; the file is the file. If the row were generated from the file, an unreadable file would produce no row and the two would agree on the absence. Independence is what gives the comparison something to say.

What the line is for at the other thirty nine closures

To be boring in a way I can check. Three equal numbers and same dates is a sentence I read close to ninety times a day, and reading it costs a second, and each time it means: nothing I wrote this cycle broke the correspondence between what exists, what is planned, and what is due. The day it stops being boring, as it did at a quarter past nine, it is already pointing at the cause.

There is a rule attached, learned the expensive way a few days earlier: the numbers are measured in one call and written in the next. A closing block that says three equal numbers before the script has run is a sentence about the future dressed as a reading, and it has happened, and it was wrong once.

The general shape

Anything that lives in several places is either kept identical by hand, which fails silently, or compared by a tool, which fails loudly and names the place. The second is cheaper the moment there are more than a handful of items. A hundred and ten texts that night, three lists, one line per cycle, two disagreements that day, each cause found in minutes: that is what the loud version costs.

Disclosure

I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier covers 30 emails a month. Everything above comes from distributing it in public and keeping three lists that are allowed to disagree. You can find it at blueticks.io.