Egypt’s Labour Law does not list Egypt’s public holidays. Article 52 delegates them — to a decision of the minister responsible for labour. The list that actually governs is a ministerial decision of 26 December 2025, issued under the new Labour Law of 2025 and a Cabinet decision from 2018.
Read that decision and several things settle at once. Eid al-Fitr is two days, not one: the decision says the first and second days of Shawwal. Eid al-Adha is three — the day of standing at Arafat and the two days following. Sham El-Nessim is the Monday after Coptic Easter, so it needs a computus and not a fixed date. The 25 January and 30 June holidays are in the decision and were missing from our record entirely. And 1 January, which we were carrying, is not in it — Egypt does not observe it.
One detail is worth pausing on. Coptic Christmas had been modelled as a Coptic calendar date, which is the intellectually satisfying answer. The decision names the day as “the Glorious Nativity (7 January)” — the Gregorian date, written out, inside the instrument that creates the holiday. So the holiday is a fixed Gregorian date, because the law says a number rather than a rule. The calendar that generated the feast historically is no longer the authority for when the country closes.
Two kinds of country
Egypt is a statute country. One instrument, amendable, citable. Retrieve it once and the calendar is settled for years; the way it goes wrong is that an amendment passes and nobody notices.
Nepal, Bangladesh, Pakistan and Indonesia are notification countries. There is no consolidated holidays statute to find. The list is reissued every year, frequently as a scan, and a retrieval buys you exactly one year.
Pakistan has no consolidated holidays statute at all — the Cabinet Division notifies the list annually, and the Cabinet’s own site carries no findable link to it. That is research before it is even capture.
The second class is not harder to find. It is harder to keep. Everyone building calendar data optimises for discovery — can we locate the source — when the property that actually governs long-run accuracy is expiry. A statute country stays right by default and fails silently on amendment. A notification country goes stale on a schedule you can predict to the day.
We can predict one: Nepal’s records run out on 13 April 2027. After that the country is blank until somebody fetches the next notification. That is written down, and it is the only reason anyone will know.
Knowing when not to publish
Bangladesh’s 2026 notification is a four-page scan with no text layer. We ran optical recognition over it in Bengali, at 300 dpi and again at 600, then applied the check that makes OCR trustworthy: the notice prints a weekday beside each date, so parse every date and test whether it lands on the weekday the document says.
Not one date passed. Month names and years came out mangled — the machine read 2028 where the page says 2026. The same method on Nepal’s Devanagari had passed all thirty-two of its cross-checks.
So the record was not written. This corpus holds six of Bangladesh’s roughly twenty-two holidays, and that is the correct number to hold, because six known days beat twenty-two of which an unknown share are wrong. The failed attempt is filed with its error rate stated, so the next person does not spend the afternoon rediscovering that Bengali OCR fails on that scan.
And sometimes the right answer is to change nothing
Brazil was on the same list and needed no work. Its two unverified national records are deliberately held at a middling grade because Brazilian law makes religious holidays a matter for municipal legislation. Good Friday and Corpus Christi sit in the national layer as a convenience, and the records say their legal basis is municipal.
The label was already telling the truth. Not every unverified row is a defect waiting for effort; some are an accurate description of a country whose law does not work the way the schema’s default assumes.