What the world's clocks rest on: an audit of the IANA Time Zone Database
收藏资源简介：
What the world's clocks rest on An audit of the IANA Time Zone Database before 1991, and three national archives read to see what is actually there. Author: Ivan Petrov Ivanov (OCO, Bulgaria) · ivan@astrocogy.com Measured against: IANA tzdb release 2026c --- Why this exists Every phone, computer and airline system on Earth shares one table of when each country changed its clocks. Its name is the IANA Time Zone Database. When your laptop tells you what time it was in Sofia, or Casablanca, or Cairo on a night in 1985, this table is what it consults. For the years before 1991, about a quarter of that table names no source at all. The database says so itself, in its own file headers: unless otherwise specified, the source for data through 1990 is Thomas G. Shanks and Rique Pottenger, *The International Atlas* — "unfortunately this book contains many errors and cites no sources." That part is not news, and this deposit does not pretend it is. The maintainers published the same assessment on their own site in 2018, and theory.html, which ships inside the database itself, puts it harder: the database "is not authoritative, and it surely has errors." Two things were missing. A number. And anybody going to look. The number Every Rule and Zone line in the geographic files that names a year between 1915 and 2005 was classified: does the comment block above it name a source — a person, a document, an assumption, another country's behaviour — or nothing? About a quarter name nothing, and fall to the declared default. No sharper figure is quoted here on purpose. The share moves depending on how you decide a comment block counts, and a number that moves is worthless unless you can re-run it. The classifier, its settings and the per-line output are in this deposit so that anyone can disagree with it precisely. Under the settings published here the answer lands between 23.6 % and 26.3 %. ⚠ Our own first pass did not survive checking. When these numbers were re-counted independently, the original headline was too high, the per-territory tier counts did not sum to their own denominator, and one sentence in the first draft — that nobody had ever gone to an archive — was simply false: the database cites hundreds of national gazette and government archive URLs. All of that is recorded rather than quietly corrected. A paper about unverified claims that does not publish its own error rate has not understood its subject. Three archives, three different problems This is the finding, and it was not what was expected. Bulgaria — the dates are there and they are wrong. Sixteen transitions between 1983 and 1990 are recorded at 02:00 or 03:00. Every Bulgarian source of the period says midnight. The end of summer time in 1996 is recorded a month late. Read in the originals at the National Library in Sofia. → https://doi.org/10.5281/zenodo.22695589 Morocco — the dates are right and there are not enough of them. Six transitions can be compared against a decree of the *Bulletin officiel*, and six of six agree, to the day and the hour — including a transition at noon and two dates fixed by a Hijri reckoning, none of which could have been guessed. What is wrong is silence: three whole summers with no rule at all, and one period that ends in a bare year instead of a date. Egypt — the dates are there and the summers did not happen. Law 87 of 1975 abolished summer time. Law 51 of 1982 restored it. Law 4 of 1985 abolished it again. Law 141 of 1988 restored it. Compile the database and read the result: Cairo is on daylight time every single year through all of it, including the nine years when Egyptian law had taken it away. And the errors do not stay put. This is why the file is worth auditing as a whole rather than country by country alone. Of Romania, europe:2485 says: > "For lack of better info, assume that Romania switched to EU rules in 1997, the > same year as Bulgaria." Romania's 1997 date is explicitly an inference from Bulgaria's. If the Bulgarian record is wrong — and the National Library says it is — then the stated basis for Romania is weakened at the same moment, without anyone touching the Romanian entry. Moldova sits behind that again: its substantive citation for 1990 is a correspondent's relay, and no Moldovan legal source appears for 1991–1996. ⭘ No claim is made about Romania or Moldova. Neither has been read in an archive. They are named because an uncited date in one country becomes a cited basis in another, and that is a property of the file rather than of any one entry. So "the atlas is unreliable" is too blunt to be useful. Wrong dates, missing dates, and dates for summers that were repealed are three different defects needing three different fixes — and nothing in the file tells you which one you are holding. The cheap test this suggests Take the transitions where the database has a date and a national instrument survives, and compare them. Agreement means the source was reading real documents, and the gaps are omissions worth filling. Disagreement means the dates themselves need review. Morocco returned six of six. Bulgaria returned wrong. It costs an afternoon per country, and it tells you which kind of problem a country has before anyone spends a month in a reading room. We have not found it run anywhere. What this is not It proposes no change to any zone. Corrections go to the IANA tz mailing list one country at a time, with the instruments attached, as Bulgaria's did. It is not an accusation. The people who maintain this database documented the weakness themselves, named the era they do not trust, and said so in the shipped documentation. They are volunteers maintaining a public good that almost nobody funds. Nothing here is a complaint about their work. Three countries is three countries. 247 territories were graded from the file alone; only three have been read in an archive. Everything outside those three is a ranking of where to look next, not a finding. Who this falls on Transitions before 1991 matter mostly to calculations made backwards into the past. The largest everyday consumer of backward time calculation is the birth chart, and the atlas was published by a company that served astrologers. ⚠ That is provenance, not causation. The atlas contains no astrology — it contains latitudes, longitudes and clock history, which are civil facts. The Moroccan result is the corrective: that same source was right six times out of six. It is noted because the loop is real and unremarked, not because it explains any error. Rights The audit, the classifier, the gradings and the analysis are the author's, released under CC BY 4.0. Reuse them with attribution. The IANA Time Zone Database and its documentation are in the public domain, and are quoted here as evidence, by file and line number, against release 2026c. The Egyptian, Bulgarian and Moroccan legal texts referred to are official state documents, which copyright law in all three countries and the Berne Convention (art. 2(4)) leave outside protection. Where a page image is reproduced it is in the country deposit for that country, identified so it can be found at source. Status Nothing here is a patch. A message describing this count is going to the IANA tz mailing list, whose archive is at https://lists.iana.org/hyperkitty/list/tz@iana.org/. This deposit is versioned. Each country read in an archive is added as a new version under the same concept DOI, which always resolves to the newest. Morocco is next, with its pages and a compiled patch.
