The jurisdiction records held a code, a name, the languages, the time zones. They did not hold a rest day. Nothing in the system knew which days of the week any country stops on, so everything that needed to know assumed Saturday and Sunday.
It surfaced in the least forgiving place available: a chart on the Saudi Arabia country page, shading Saturday and Sunday as the weekend, on the page a reader opens precisely to find out that the Saudi working week runs Sunday to Thursday.
We removed the band rather than guess at it. But the chart was the symptom.
What rests on knowing where the weekend is
Almost everything, once you look:
- Long weekends. Every “four-day run” and every “take the Friday off” on 241 country pages was computed against a Monday–Friday week.
- Working-day counts. The “261 days in the year” figure, everywhere it appears.
- The calendar shading across the explore view, which greyed Saturday and Sunday for every jurisdiction on earth and told the reader the weekend “already answers no”.
- The working-days, long-weekends and deadline tools, all three.
For roughly twenty Gulf and Islamic-calendar jurisdictions, each of those was wrong in the direction that costs money — telling you an office is open when it is shut, or offering you a long weekend that is an ordinary one.
The irony worth recording is that the corpus already handles the harder version of this problem. China’s 调休 arrangement, where a Saturday is worked to bridge a holiday, is captured properly. We could describe a country working through its own weekend, while being unable to say which days its weekend fell on.
Why it cannot be typed from memory
The fix is small — one field, 241 rows, and labour codes and central-bank working-week notices state it plainly. It is sourceable exactly like everything else here.
What it is not is recallable. Several states have changed within living memory: Saudi Arabia moved from Thursday–Friday to Friday–Saturday in 2013, Oman the same year, and the United Arab Emirates moved to a half-day Friday in 2022.
A remembered answer here is likely to be a former answer — and confidently so, because the person remembering learned it once and had no reason to check again. That is the general hazard with any fact that felt permanent when you learned it.
The shape of the failure
This is not a wrong value. It is a missing field, and a default standing in for it so quietly that nobody saw a decision being made.
Nothing was malformed. No check could fire, because there was no field to validate and no value to compare. The system was internally consistent and externally wrong for about a tenth of the world, and it took building a picture of a country — the one deliverable where a human has to look at the result — to notice.
Defaults are the most expensive thing in a dataset, because they never announce themselves. A null shows up in an audit. An assumption is indistinguishable from a fact until something forces it into the open.