Thailand's Royal Gazette set a record in 2025, and nearly 6 in 10 notices are bankruptcy cases

The Royal Gazette (ราชกิจจานุเบกษา) is where Thai laws, notices and orders take legal effect. Almost nobody knows what goes into it each day, because the official site lets you search one notice at a time and gives no overview.

In late 2025 the Open Law Data Thailand project released every Gazette notice since 1885, about 1.39 million of them, as an open dataset on Hugging Face, with full text and category labels, updated every working day. We downloaded the label layer and the per-notice metadata layer in full and counted them by script. The numbers here run to 25 September 2026.

This article shows no names of people who appear in notices, and it is not legal advice.

2025 published more than any year on record

The 49,127 notices of 2025 beat every other year in the dataset. The closest are 2010 (47,593) and 2011 (46,663).

The other clear pattern is that the Gazette is heaviest on Wednesdays. In 2025, 19,401 notices came out on a Wednesday, about 4 in 10 of the year, against 4,612 on Mondays and almost none at weekends (143 on Saturdays, 51 on Sundays). If you follow a particular kind of notice, Wednesday is the day to check.

Nearly 6 in 10 notices are bankruptcy cases

Receivership orders, bankruptcy judgments and discharges must be published in the Gazette case by case, so this group is most of what the Gazette carries on any given day.

Monthly totals in 2026 range from 1,815 (May) to 2,764 (February), with 2,758 in September. These count notices, not people: one case can produce several notices as it moves through its stages.

The dataset also records whether the debtor is an individual or a company, but we can only use that from April to September 2026, because before then a large share of notices had no type extracted (see the limits below). In those six months, of 13,221 notices, 12,341 or 93% concerned individuals and 858 concerned companies.

These notices carry the debtor’s full name in the title. Even though the state has already published them, turning them into a web page searchable by name is a different matter, so we show totals only.

What the other 4 in 10 are

The next largest categories in 2025, counting only labels the dataset marks as confirmed:

Local public health made the list in 2025 for a specific reason, which turned out to be the most interesting thing we found.

In early 2025 the Gazette published a backlog of over 3,000 local ordinances

Notices from local governments (subdistrict administrative organisations, municipalities and provincial organisations) normally run at about 130–300 a month. Most are ordinances on everyday matters: controlling the keeping and release of dogs and cats, regulating markets, food premises, grease traps in buildings, and businesses that pose a health hazard.

In the first three months of 2025 the numbers jumped:

Local public health ordinances alone came to 3,027 notices from at least 1,710 differently named local bodies in those three months (names repeat across provinces, so the true number of bodies cannot be counted).

What shows this was a backlog is the year in the ordinance’s own title. Almost every ordinance names the year it was enacted in its title, in the form “B.E. 25xx”. We compared that year with the year of publication.

The ordinances published in early 2025 had been enacted two to three years before. From April 2025 on, 8–9 in 10 ordinances were published within a year of enactment, and that is still the case in September 2026, except for July 2026, when 36% were old ones. (We use “within a year” rather than “the same year” because ordinances enacted late in a year are usually published early in the next.) The data shows what happened, not who was holding the backlog. We found no document explaining it, so we do not guess.

One side effect: local-government notices totalled 5,420 in 2025 but only 1,035 in 2026 up to September (with one smaller bump, 375 in January 2026). Read the yearly figures alone and you would wrongly conclude that local governments are making far fewer rules.

Data limits to know before you use it

• The province field is almost empty. The dataset has a province field, but only 1 of 122 local-government notices in September 2026 has it filled. The organisation’s name cannot stand in for it, because the Department of Local Administration’s official 2026 list of 7,841 local bodies (Bangkok excluded) has 619 names that repeat across provinces; there is a subdistrict organisation named Nong Bua in 19 provinces. The full text names a province in only 3–4 notices out of 10 (32 of 79 in August, 35 of 122 in September).
• Debtor type is reliable only from April 2026. From mid-2024 to March 2026, 35–74% of bankruptcy notices per month had no type extracted, above all the 2025 “adjudged bankrupt” notices, where 8,862 of 8,927 had none. From April 2026 the figure is 0%.
• Some notices have no full text. 5,146 notices from 2021, 4,289 from 2003 and 3,455 from 2004 have none. Some were withheld by the project itself because it could not prove the file matched the document code. Their category labels come from the title alone.
• Labels come in two grades. The project reports that labels backed by structural evidence were 100% correct in a 450-notice human check, while unconfirmed labels are about 61% accurate. Topic labels are confirmed for 94% of notices, but the “what action” labels (rulemaking, appointment, registration) are confirmed for only 24%. We counted confirmed labels only and did not use the action labels as headline figures. We also read the titles of a random 24 notices across 6 categories; the label matched the title in all 24.
• Labels start in 2000, but for 2000–2001 they come from titles alone and the project itself does not count those two years as covered. Earlier notices have per-notice metadata but no labels.
• The licences do not agree. The notices themselves are not copyrighted under section 7 (2)–(4) of the Copyright Act B.E. 2537, which covers laws, notices and judgments. For the layers the project added, the Hugging Face page says CC BY 4.0 and the project website says CC BY-SA 4.0. We credit the project and link back, and we do not distribute modified files.

How we counted

• Downloaded the taxonomy layer (812,582 notices) and the meta layer (1,390,468 notices) from Hugging Face.
• Always used the publication date (publish_date), never the year in the file name, because files are organised by document code.
• Counted only labels where topic_corroborated or govlevel_corroborated is true.
• For the ordinance lag, used only local-government notices whose “rulemaking” label is confirmed and whose title contains a Buddhist-era year (“พ.ศ. 25xx”); see measure_article_a.py.
• Checked the update time from the dataset’s commit history: it updates on working days between about 21:00 and 23:00 Bangkok time, so each day’s notices arrive the same evening.
• The scripts are kept in this site’s code, so anyone who downloads the same dataset can rerun them.

Sources

• The soc-ratchakitcha dataset on Hugging Face — README, commit history, and the taxonomy and meta files
• Open Law Data Thailand — project page and licence
• Royal Gazette — original PDFs for legal citation
• List of local administrative organisations 2026, Department of Local Administration — used to check names repeated across provinces
• DGA-Thailand/open-law-data on GitHub — code and showcases built on the dataset