A record of errors found after publication and how we corrected them, including what we got wrong and why.
Repositioned related reading and clarified bibliographic links and the manipulation exception.
Why we got it wrong Consolidated overlapping research in the existing article and clarified the provision and related reading.
Separated legal categories from nonprofit verification, clarified possible document requests, and removed guaranteed CTR outcomes or fixed improvement periods.
Corrected claims about unincorporated groups, nonprofit requirements, Maps credits, and YouTube donation eligibility.
Why we got it wrong Checked primary guidance from the National Tax Agency, Google, and Japan Post Bank; fixed timelines and annual benefits were unsupported.
Corrected blanket claims about registration, timing, governance, product grants, and nationwide inhabitant-tax exemptions.
Corrected notarization, current NPO inspection timing and receipts versus taxable income; removed claims that separate entities alone prove savings or business independence.
Clarified which bidding strategies and accounts the CPC and CTR rules apply to.
Why we got it wrong Updated against Google's operating guidance and account management policy. https://support.google.com/nonprofits/answer/98870 https://support.google.com/nonprofits/answer/117827
Removed unsupported estimates of Japanese Ad Grants adoption and claims about the main causes of nonuse.
Why we got it wrong The Google sources checked report a worldwide cumulative count, not a Japanese user count or a survey of reasons for nonuse. Comparing the NPO corporation count with an unsupported Ad Grants estimate cannot establish adoption. https://www.google.com/grants/program/
Corrected fixed page and word counts, yen conversions, and the claimed start year for conversion tracking.
Why we got it wrong Google's current website policy requires substantial original content but states no five-page or 300-word minimum. The conversion help does not give 2025 as the start year. Local-currency conversion fluctuates and the budget limit is not guaranteed spend. https://support.google.com/nonprofits/answer/1657899?hl=en https://support.google.com/nonprofits/answer/9841491?hl=en https://support.google.com/nonprofits/answer/1332166?hl=en
Removed the automatic-benefit claim and the ROI calculated from an advertising budget ceiling.
Why we got it wrong Google's budget guidance does not guarantee usage. No organization-specific approval or outcomes were verified.
Separated the Ad Grants budget ceiling from for-profit savings and corrected the two Japanese non-profit tax categories.
Why we got it wrong Reviewed Google's budget, mission and website policies and the National Tax Agency's category-specific requirements. A full tax or performance audit remains outstanding.
Distinguished the Ad Grants budget ceiling from actual advertising cost savings.
Why we got it wrong Google does not guarantee that an account will spend its full budget. A currency conversion of the ceiling does not establish actual cost savings. https://support.google.com/nonprofits/answer/1332166?hl=en
Corrected unsupported fixed website and conversion timing rules.
Why we got it wrong Google's current website policy does not set a minimum number of pages or words. Its conversion tracking help does not state that the rule began in 2025 or that every account is suspended after 60 days without conversions. https://support.google.com/nonprofits/answer/1657899?hl=en https://support.google.com/nonprofits/answer/9841491?hl=en
Aligned the denial and new-request guidance with Google's current help page.
Why we got it wrong Google's current help page does not state a universal waiting period or a separate 30-day review estimate for a new request. Checked on 2026-09-29. https://support.google.com/nonprofits/answer/2457150?hl=en
Corrected the application entry point and distinguished verification from overall review timing.
Why we got it wrong Google's current help says it shares the application with Goodstack. It does not require a separate Goodstack-first account and code in the published process. The Goodstack and overall review estimates do not guarantee activation time.
Removed the fixed yen conversion and the universal three-month wait after denial.
Why we got it wrong Google's current help says local-currency conversion fluctuates and the monthly limit is not guaranteed spend. Its new-request instructions do not state a universal waiting period. Checked on 2026-09-29. https://support.google.com/nonprofits/answer/1332166?hl=en https://support.google.com/nonprofits/answer/2457150?hl=en
Aligned the application entry point and review timeline with Google's current instructions.
Why we got it wrong Google's help pages say Google sends application details to Goodstack. Three to five business days describes review of most Goodstack requests, while the overall Google for Nonprofits guidance is two to fourteen business days. Neither guarantees approval or product activation.
Updated the companion piece's title to its new title, and corrected the passage that described the companion piece as being about cultural budget cuts.
Why we got it wrong The companion piece's title was corrected on 26 September 2026. The Agency for Cultural Affairs' initial budget was almost flat from FY2024 to FY2026, and the companion piece itself does not say that Japan's cultural budget was cut, so this article's description of it now matches. Its links now also point to the English version.
Replaced the number of public libraries and the materials figures from the 2020 tabulation with those from the 2025 tabulation.
Why we got it wrong The figures added in the correction of 25 September 2026 came from the Japan Library Association's 2020 tabulation, and the text did not give the year. In the year-by-year table published in March 2026, which runs to 2025, the number of libraries is almost flat and the budget for library materials has risen. The reason given for the 25 September correction, that the number of libraries had risen and that the only decline in the figures checked was a slight fall in materials spending, was also based on the 2020 tabulation.
Brought the title, the image headline for social media and the description in line with what the text establishes.
Why we got it wrong The Agency for Cultural Affairs' initial budget was almost flat from FY2024 to FY2026, and so was the number of public libraries, which does not fit “closing libraries” or “cultural budget cuts”. The text notes that some countries invest in both, and does not say that culture is cut to pay for defense, so “tradeoff” was dropped as well. The ¥106 billion is the Agency's budget; “for culture” reads as all spending on culture, so it now names the culture agency.
Changed the title, the image headline for social media and the description to refer to the Agency for Cultural Affairs' budget rather than to culture or the cultural budget.
Why we got it wrong The roughly 0.02% is the Agency for Cultural Affairs' initial budget alone divided by nominal GDP. When the text was corrected on 25 September 2026, the title, the image headline for social media and the description still referred to culture or the cultural budget, which reads as the share of all spending on culture by the state and local governments combined.
Brought the summary shown under the Background & Context heading in line with what the section covers.
Why we got it wrong The section covers the designated manager system, free admission abroad and cultural rights, and does not trace a history of underfunding. The Japanese summary had also spoken of budget cuts, although the Agency for Cultural Affairs' budget rose from ¥106.3 billion in FY2025 to ¥107.3 billion in FY2026.
Brought the description of the medium-term objectives for national museums in line with the original text and removed the references to closure and zero operating grants.
Why we got it wrong The original objectives do not mention closure, and the Agency for Cultural Affairs page this paragraph linked to states that reorganization does not envisage closing museums and that a share below 40% alone does not immediately make a museum subject to reorganization. The 65% and 100% figures apply only to exhibition programs, and the National Museum of Art's objectives describe securing appropriate operating grants as indispensable. The source for the figure now points to the objectives themselves instead of a news article.
Brought the descriptions of the San Diego study, the research covered in an NEA blog post, the IMLS budget, UK library spending cuts and closures, and the Ministry of Education report in line with their sources.
Why we got it wrong The authors and year of the paper were wrong, and the figure of 150 people came from the San Francisco example the paper cites, not from San Diego. The 69% and 93% also came from 49 patrons experiencing homelessness or housing instability and 14 staff, not from library patrons in general. What the text called NEA research was a study published by the IMLS and described in an NEA blog post; it found no significant relationship for museums. The IMLS budget figure does not appear on the ALA page it was attributed to; the IMLS budget table gives $294.8 million for FY2024. The UK source puts the cut at one-quarter to three-eighths of expenditure and gives no 50% figure, and the closure figures now cite the BBC Data Unit. The Ministry of Education page does not contain the quoted phrase. The budget document and the book title in the references were also aligned with the originals.
Corrected how the defense budget's share of GDP and its growth were described.
Why we got it wrong The figure of roughly 1.8% of GDP does not appear in the Ministry of Defense document cited. Press reports in April 2025 gave it for the government's total of about ¥9.9 trillion, which adds items such as the Japan Coast Guard budget and public works spending usable for security to the defense budget, so it is not the ratio for the ¥8.7005 trillion of defense-related expenditure. The claim of about ¥1 trillion a year did not match the latest one-year increase of about ¥750 billion, so it now uses the difference between the Ministry of Defense figures for fiscal 2022 (¥5.4005 trillion) and fiscal 2025 (¥8.7005 trillion).
Replaced the claim that Japan's cultural infrastructure is being cut with what the sources support.
Why we got it wrong This section presented an existential threat to national museums and shrinking public libraries as examples of Japan's cultural infrastructure being cut. After the earlier correction removed the threat, the sources showed that the Agency for Cultural Affairs' initial budget is roughly flat and that the number of public libraries rose from the previous year; of the figures checked, only materials spending fell, and only slightly. The text therefore no longer says that Japan is cutting cultural infrastructure. It says the budget has not grown as defense spending has.
Corrected how UK library spending cuts, the culture and leisure share of council spending, and the fall in library numbers were described.
Why we got it wrong The source, Public Libraries News, says spending is £232.5 million (£329 million with inflation) lower than in 2010, not a cumulative total of yearly cuts. The 7.4% and 4.5% come from the Local Government Association (LGA), the councils' representative body, as the share of service spending excluding education, public health, police and fire. A fall from 7.4% to 4.5% does not mean the share was halved, so 'effectively halved' was removed. In the BBC Data Unit's material the 150 is a net figure since 2016 and about 100 of those libraries continue under community groups, so the wording about losing 150 in a decade was dropped. The heading and key point that still said closure rates were four times higher now match the body's 'four times more likely to have lost a publicly funded library'.
Corrected how IMLS staffing, the library associations' statement and the lawsuit were described.
Why we got it wrong No source could be found for 'all but 12'. The ALA and the American Alliance of Museums said that the entire IMLS staff had been placed on administrative leave, so the text now says nearly all. The linked statement was issued jointly by the Public Library Association (a division of the ALA), the Association for Rural & Small Libraries and the Urban Libraries Council, not by the ALA alone. The account of the lawsuit now cites the ALA announcement of 7 April 2026.
Brought the description of the Sixth Medium-Term Objectives in line with the original text, corrected what the 65% is a share of, and removed the references to closure and zero operating grants.
Why we got it wrong The 65% is not a share of total income but the share of exhibition costs covered by self-generated revenue, and the objectives were set by the Ministry of Education, Culture, Sports, Science and Technology, not by the Agency for Cultural Affairs. The original objectives do not mention closure, and the Agency for Cultural Affairs has stated that reorganization does not envisage closing museums. The 65% and 100% figures apply only to exhibition programs, and the objectives describe securing appropriate operating grants as indispensable.
Replaced the international comparison chart and text with the figures from the Agency for Cultural Affairs' comparative survey.
Why we got it wrong The GDP shares in the chart had no source, and the Agency's comparative survey gives no GDP shares at all. The chart and text now use the share of government budget the Agency publishes, and include the United States, which the same survey covers. Japan's figure of about 0.02% of GDP is now marked as a calculated figure.
Corrected the year of the Tokyo National Museum's fee rise, the Louvre's prices, free admission in France and Germany, the operating grants, curators, the Basic Act on Culture and the Arts, and the National Museum of Nature and Science's crowdfunding, in line with the sources.
Why we got it wrong The Tokyo National Museum's fee rise took place in 2020. No sources were found for the special exhibition prices or for the curators' salary and workload figures. The claim that grants were declining every year contradicts the Agency's statement that they rose in FY2026. The Louvre and the free admission rules in France and Germany now follow the official pages. The text of the Act does not use the words “social inclusion”. The supporter count and the record now follow the Nikkei article cited. The budget document in the references was also updated to the same year as the text.
Made clear what the 0.02% counts and which year the international comparison covers, and added spending that falls outside it.
Why we got it wrong The roughly 0.02% is the Agency's initial budget divided by nominal GDP, not spending on culture by the state and local governments combined. The tourist-tax projects and supplementary budgets sit outside the initial budget, and the Agency's comparative survey puts local governments' cultural spending in Japan at ¥470 billion. The comparison figures are for 2020, but the year was not stated.
Replaced statements without sources, on a rise in museum closures, the casualization of curators and a principle behind free admission in the UK, with content that has a source.
Why we got it wrong No statistics showing that museum closures are rising could be found. The Ministry of Education, Culture, Sports, Science and Technology's Survey on Social Education (FY2021) counted a record 1,305 museums, and museum-like facilities also rose from the previous survey, to 4,466. The Japanese Association of Museums' FY2024 survey says non-regular staff are about a quarter of all staff and that their increase has been halted. No source was found for curators being on three-to-five-year contracts, and the wording of the UK principle could not be confirmed, so the text now gives only the start date that could be confirmed.
Rewrote the paragraph on the erosion of publicness through self-revenue demands, limiting it to exhibitions and basing it on what the medium-term objectives say about admission fees.
Why we got it wrong The 65% applies only to exhibitions, and the Agency says the state budget funds the other three areas, so the pressure cannot be described as one that turns museums as a whole into revenue-generating enterprises. The rise in admission fees and the introduction of dual pricing, however, are in the text of the objectives. The National Museum of Art's announcement says free admission to the collection exhibitions for high school students and younger, under-18s, those aged 65 and over, and holders of disability certificates with one companion continues, so the elderly and people with disabilities were removed from the groups said to be excluded.
Added to the comparison with other countries that Japan's national museums also have free admission groups for collection exhibitions, and that the Louvre's prices differ by place of residence.
Why we got it wrong Free admission for under-18s exists at Japan's national museums as well. A comparison that lists only the arrangements abroad makes the gap with Japan look larger than it is, so this was added.
Corrected the organization named in this note. The group we described was linked to a different body, and we removed a founding year that the linked page does not state.
Why we got it wrong The Japanese Society for History Textbook Reform is the body that English Wikipedia itself dates to December 1996, and it is not the group this note discusses. The phrase "established in 1995" did not appear on the page we linked to, so we removed it.
Dropped the claim that this is the first attempt of its kind.
Why we got it wrong No source was checked for the absence of earlier attempts.
Replaced the claim that South Korea's AIDT downgrade had no precedent anywhere in the world with what the sources support.
Why we got it wrong No source was checked for the absence of comparable cases in other countries. The introduction and downgrade dates and the five-month span are in the cited source, so only those were kept.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Dropped the claim that a mid-cycle revision is virtually unprecedented since 2000.
Why we got it wrong No source counts the revisions since 2000. The cited research institute calls it a mid-cycle revision but says nothing about precedent. The argument that the government overrode its own three-year cycle is carried by the sentences that follow.
Matched the English wording for the 66-yen increase to the Japanese, which says largest on record.
Why we got it wrong The Japanese version says largest on record, based on the ministry materials and reporting cited in the body. Only the English had been translated up to unprecedented.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Corrected the yen conversion of the Ad Grants ceiling, which was out by a factor of ten.
Why we got it wrong The cap is up to $10,000 a month, which comes to $120,000 over twelve months; at 150 yen to the dollar that is about 18 million yen, not 1.8 million. (1.8 million yen is what a tenth of the cap comes to.) The yen figure is our own conversion, not one Google publishes. The dollar figures on this page already read as a ceiling, so they are unchanged.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Dropped the word unparalleled.
Why we got it wrong No source was checked for the absence of comparable programs. The amounts and the maintenance requirements remain sourced in the body.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Changed "continue to occur" in the description to the count the article actually gives. The link to a related note now uses its current title.
Why we got it wrong The article cites three cases, in 1974, 2002 and 2021. Calling three cases across 47 years an ongoing occurrence is a claim about frequency that the article does not support. The link label was left behind when we removed "world-first" from the target note's title on 2026-09-14.
Removed a sentence that the source does not contain, and pointed the 43.6% citation at the document that carries the figure instead of the survey's landing page.
Why we got it wrong We opened both published FY2023 documents, the summary and the report on living and management, and neither links year of completion to residential-noise disputes. The one figure broken down by year of completion is arrears on management fees, where older buildings show the higher rate, the opposite direction. The 43.6% itself appears on page 28 of the summary as "residential noise at 43.6%, the most common," so the figure and its framing stand.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Rewrote the Ad Grants figures so they read as a cap.
Why we got it wrong Google's own page says "up to $10,000 per month." The amount is a ceiling, not a grant paid out in full each month, and writing "$10,000/month in advertising" reads as money received. The $120,000 a year figure is what the cap comes to over twelve months if it is used in full.
Corrected the yen conversion of the annual advertising value, which was out by a factor of ten.
Why we got it wrong $10,000 a month over twelve months is $120,000, which at 150 yen to the dollar comes to about 18 million yen, not 1.8 million. (1.8 million yen is what a tenth of the cap comes to.) The yen figure is our own conversion, not one Google publishes.
Replaced the claim that the 1 trillion yen scale was unprecedented with the figure itself.
Why we got it wrong No source compares this against earlier investment in people. The figure itself is stated in the body and stays.
Attributed the world-first wording to the New Zealand Treasury instead of asserting it.
Why we got it wrong World's first is the New Zealand Treasury's own description, so the article now says whose description it is.
Replaced a pace unprecedented in history with what the article's own figures support.
Why we got it wrong No source compares this against the pace of past structural transformations. The 24 years remaining and the 30 to 50 years such shifts usually take are both in the body, so the sentence now says only what those figures support.
Replaced unprecedented confrontation with unusual confrontation in the description and the summary.
Why we got it wrong No source was checked for the absence of comparable confrontations. The lawsuit itself and its contents are sourced in the body and stay as they are.
Replaced "world's first comprehensive law" with what the source supports, "one of only eight countries out of 194," in the title, summary, section summary, social copy and body. Added the missing source for the figure of eight countries.
Why we got it wrong The Cabinet Office page for the Act carries no "world's first" wording, and searching for the two together returns no source. The figure of eight countries, by contrast, comes from a review of all 194 WHO member states, and the country list matches the article. The claim that Japan's is the only one in the form of a comprehensive law is not in that review, so it has been dropped.
Replaced "never intersect" and "complete non-intersection" with what our literature searches can support. The description and the social post now use the same wording.
Why we got it wrong Two paragraphs below that sentence the article already says the absence is within what these searches reach, not a claim to have exhausted the world's literature. Only the bold line overstated it, and we missed it when we corrected the same kind of wording on 2026-09-14.
Replaced unprecedented tension with sharp tension in the body, the section summary and the figure caption.
Why we got it wrong No source was cited for the comparison against earlier federal-state conflicts. The article itself also says, in another figure caption, that this conflict is not specific to AI and repeats a structural pattern seen in past cases such as environmental regulation, so the two statements contradicted each other. The figure of 45 states introducing bills is sourced and stays.
Replaced the "world first" wording for H1 with what our literature searches can support. The research-gap table and the link to the related note now use the same wording.
Why we got it wrong Calling an absence in our searches a world first claims that we checked the world's literature. We missed this overview note when we corrected the same wording on the individual note on 2026-09-14.
Applied the same wording to H2 through H4, which were still left in this note.
Why we got it wrong Earlier the same day we corrected only H1 and missed the other three lines in this same note. "Japan first" and "not implemented" both state an absence in our searches as an absence in the world.
Dropped the claim that the scale and speed of the cuts are unprecedented.
Why we got it wrong No source was cited for the comparison of scale and speed against past retrenchments. The single-bill form and the size of the cuts are sourced to CBPP in the body, so only those were kept. The English version also stated this more strongly than the Japanese; both now match.
The scholar named in the text and the book listed in the references belonged to two different people. We removed both and replaced them with the government guidelines on public records management.
Why we got it wrong The book listed in the references was written by Kazuo Yano, not by Taku Sugawara, whose name appeared in the text. The book itself reports research on happiness and productivity measured with wearable sensors; it does not discuss the keeping of public records. When writing the essay, we checked neither the author nor the content of the book.
We replaced the figure 'more than 300 passages' with '14 approval documents', the number stated in the Ministry of Finance document, and removed the sentence about the purpose of the rewriting.
Why we got it wrong The report we cited is a scanned file whose text cannot be read by machine, so we could not confirm where the figure 'more than 300' appears in it. A separate document the ministry published on the same case states that 'a total of 14 approval documents were rewritten', and that text is machine-readable. We could not find the statement about the purpose of the rewriting in the cited source either.
Attributed the definition of civic media to the book that actually gives it (Civic Media, edited by Gordon and Mihailidis), and added that book to the references.
Why we got it wrong The definition comes from the volume Gordon and Mihailidis edited in 2016, not from the handbook we had credited. Our wording had also dropped "produce and reproduce" from the original, so we restored it.
Corrected the basis for the sample size in the validation design and the description of the WHO guidelines, and removed "the first" from the sentence about showing the threshold with measured data.
Why we got it wrong Recomputing the stated conditions (d=0.8, α=0.05, 80% power) does not yield 15-20 per group; it yields 26. The WHO guidelines themselves describe systematic reviews of the relationship between noise and health outcomes, so "demonstrate causal relationships" overstated what the source says.
Removed one reference whose existence we could not confirm (Stehr, The Uncertainties of Knowledge). The text of this essay does not cite it.
Why we got it wrong Neither Crossref nor Open Library lists a book with this title. The author's books from 2003 to 2005 include The Governance of Knowledge, Biotechnology and Society and Knowledge, none of which matches. When writing the essay, we did not confirm that each reference exists.
In the section "It Is Not a Think Tank", replaced the cited paper with an existing work, and removed the claim that Stone framed this "in the context of the IPPA", which the source cannot confirm.
Why we got it wrong Crossref lists no paper with the cited title, and the link we had attached pointed to a different paper in a different journal (Hawkins and Holden, 2014). When writing the note, we did not open the source to check it.
Corrected the authors and descriptions of two closely related prior studies to match the linked papers, and replaced the "world-first" wording in the title and description with what our literature searches can support.
Why we got it wrong When writing the note, we did not open the linked papers to check their authors and methods. Checking them against Crossref and the full texts showed that the authors, methods and study location differed from our descriptions. The "world-first" wording restated "not found in our searches" as "does not exist anywhere"; we changed it to match what the searches re-run on Europe PMC on 2026-09-03 can support.
Corrected the unemployment insurance column for all four years and removed the total column. The 9.3% figure and the 6-percentage-point rise that were derived from it have been replaced.
Why we got it wrong The employee share of unemployment insurance is half the rate for unemployment benefits. The MHLW series gives 11 per 1,000 for fiscal 1990 and 12 per 1,000 for fiscal 2000 and 2010, and the fiscal 2025 notice gives 5.5 per 1,000; none of the four figures in the article matched. We also removed the total column. Only the 1990 total came from a Cabinet Office tax commission chart, whose note states that rates before the total-remuneration system were calculated by assuming annual bonuses equal to three months of salary. A bonus-adjusted figure and a plain sum of monthly-salary rates were sitting in the same column, which is why the 1990 row alone added up to 11.75% against a stated total of 9.3%. The article now states plainly that the health and long-term care rates are not currently traceable to a primary source.
Dropped the headline 22.6% capture rate and the international comparison table, and replaced them with the protected-household ratio the ministry actually publishes.
Why we got it wrong Neither 22.6% nor 43.3% appears in any of the seven sources the article cited. The actual document, MHLW's "Estimation of the number of low-income households below the welfare standard" of 9 April 2010, calls the measure the protected-household ratio, not the capture rate, and gives 15.3% (income only) and 32.1% (assets considered) on the 2007 Comprehensive Survey of Living Conditions, and 23.8-29.6% and 68.4-87.4% on the 2004 National Survey of Family Income and Expenditure. The same document states that the number of low-income households below the welfare standard does not represent households that are eligible and intend to apply but are not receiving assistance. The three countries in the comparison table had no URLs, and we could not confirm that their figures are defined the same way as the protected-household ratio, so the table was removed entirely.
The social copy was still on the figures published a month earlier.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The FY2021 share was still the old figure in the key points.
Why we got it wrong The source publishes FY2021 by age band (0.3% under 20, 11.1% for 20-29, 30.1% for 30-39); 41.5% is the sum of those. The body was moved to the banded figures on 2026-08-18.
A share the body itself flags as inconsistent was stated as fact in the summary.
Why we got it wrong The body notes that two Children and Families Agency documents disagree on the share (185 of 504 is 36.7%, but the budget request gives 37.7%) and uses the counts instead. Only the summary still gave 37.7% as fact.
The per-prefecture shares dropped from the Japanese version were still standing in the English one.
Why we got it wrong Shares by prefecture would have to be computed by combining columns of the registration summary, so the source does not publish them in that form. The Japanese version dropped them on 2026-08-18; the English version kept them with a source attached.
Corrected the construction date of Gasholders London to that of the frames now standing
Why we got it wrong Historic England's list entry 1464325 records that the frames standing today were built in 1879–1880 to the design of John Clark, replacing the earlier frames. The article described a replacement as an enlargement.
Replaced the decade range for the Vienna gasometers' decommissioning with the actual year
Why we got it wrong The cited source states they were used from 1899 to 1984, so the year is known.
Replaced the sources for TANK Shanghai and Gasholders London with documents that state the facts
Why we got it wrong The previous sources were landing pages carrying none of the figures the article cited — apartment count, listing grade, opening date or area. They have been replaced with documents that state those figures.
Repointed the link to a law that exists and dropped the month of the amendment.
Why we got it wrong The law ID in the link does not exist on e-Gov; the API returns a 404. The Act on Promotion of Global Warming Countermeasures is Act No. 117 of 1998. The month has been dropped because passage and promulgation fall on different dates and only the year could be confirmed.
Repointed the municipality count and population coverage to the survey that actually publishes them.
Why we got it wrong The Research Institute of Local Government page carries neither 459 nor 85%. It shows only a current figure, which read 532 bodies as of 31 May 2025 when checked in September 2026. The 459 and 85.1% come from the Shibuya Ward and Nijiiro Diversity joint survey, so the citation now points at a document reporting that survey. The table's "exceed 460" also disagreed with the 459 in the body, and the two now match.
Corrected the name of the act in the references to the title e-Gov gives.
Why we got it wrong The act's title is the Act on the Settlement of Environmental Pollution Disputes. The name was checked against e-Gov law search.
The staged expansion of the substance list was still the old chain in the title, summaries and social copy.
Why we got it wrong The 896 / 1,600 / 2,900 chain could not be found in the sources and was dropped from the body on 2026-08-18. It survived in ogLead, the description, the key points and the social copy — five places in each language.
The 2026-08-18 correction had reached the body only; the summaries, social copy and household-model paragraph still carried the old figures.
Why we got it wrong The body had already been repointed at the FY2022 Prefectural Economic Accounts, but ogLead, the key points, the social copy and the household-model paragraph were left on the old numbers. Both language versions were in the same state.
Added the missing "whichever is higher" to the second and third fine tiers, and widened the second tier to the scope the article actually sets
Why we got it wrong Article 99 of the EU AI Act applies "whichever is higher" to all three tiers; it was missing from the second and third rows. Article 99(4) is also not confined to high-risk AI obligations: it covers the duties of providers, importers, distributors, deployers and notified bodies, and the transparency duties in Article 50. A source carrying the text of Article 99 has been added.
The cooperative count and the Specified Skilled Worker figure were still the old ones in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The yen conversion of Seoul's budget was still in the key points and social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The private-fee band and the post-birth income drop were still the old figures in the summaries, social copy and a heading.
Why we got it wrong The banded fee figures were dropped from the body on 2026-08-18 because the source is a paywalled article whose text could not be checked. They survived in the summaries, the social copy and a heading. The 80,000 yen used in the household model is now labelled as a working assumption.
The sum of the top three violations did not read as a figure computed here.
Why we got it wrong The counts and shares are added up here from the breakdown the NPA publishes; the totals themselves are not published.
The cumulative education-fund trust figures and the per-contract amount were still the old ones in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The FY2028 child-support levy rate was still the old figure in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Recomputed the 2019 increase and growth rate from the published MEXT figures.
Why we got it wrong The increase column carried the following year's figure (14,855) instead of its own. MEXT records 164,528 for 2018 and 181,272 for 2019, a difference of 16,744 and a growth rate of 10.2%. The rate had been computed from the wrong increase.
The highest and lowest teacher-training rates were still the old figures in the social copy, with prefecture names the source does not give.
Why we got it wrong The body was corrected to 99.6% and 56.8% on 2026-08-18, and the source names no prefectures. The social copy and the English section summary kept the old figures and the names.
Corrected the act number for the Code of Civil Procedure and the link behind it.
Why we got it wrong The Code of Civil Procedure is Act No. 109 of 1996. Act No. 109 of 1948 is the Local Finance Act. The law ID in the link did not resolve either, so it has been replaced with one verified against the e-Gov API.
Removed the yen conversions from the international comparison and sourced each country's figure.
Why we got it wrong The three yen figures implied three different exchange rates (about 147 to the dollar, 186 to the pound, 173 to the euro) and were all computed here. We fetched the full text of the 2018 Cabinet Office tax commission document cited for the table: apart from France's 15-year clawback, not one of the figures appears in it. The US figure now cites the IRS and the UK figure GOV.UK. We also noted that Germany's 30% is the ceiling for class I, which covers spouses and children.
Matched the denominator of the older generations' asset share to the one used in the source.
Why we got it wrong 1,257 divided by 2,195 is 57.3%, not the 62% the article gave. Re-reading the source shows that Dai-ichi Life Research Institute uses the survey-comparable 2,035 trillion yen as the denominator, and 1,257 over 2,035 is about 62%. Both figures are in the source; only the way they were joined was wrong.
The tower-condominium case amounts were still stated as combined totals in the key points.
Why we got it wrong The judgment's figures are totals across two properties; the body was corrected to the per-property breakdown on 2026-08-18 and only the key points kept the totals. The English summary's framing of 1,257 trillion out of 2,195 trillion has also been brought into line with the body.
The Okinawa annual average did not read as a figure calculated here.
Why we got it wrong Okinawa ran 61 days at the old 952 yen and 304 days at the new 1,023 yen, giving a twelve-month average of about 1,011 yen. Neither figure is published; both are calculated in this article.
Added the fee for a certified copy of the articles to the cost breakdown and corrected the total.
Why we got it wrong Notarization carries a separate fee for the certified copy of the articles (Notary Fee Order, Article 40(2), "250 yen per page"). It was missing from the total, so the figure understated what a founder needs to have ready. Each line now carries its statutory basis.
The Kyoto City vacancy figures dropped from the Japanese version were still standing in the English one.
Why we got it wrong In the national results cited, 14.1% belongs to Nara, and that document carries no city-level breakdown. The Japanese version dropped the city-level figures on 2026-08-18; the English version kept them.
A starting-salary comparison that had been dropped was still in the key points.
Why we got it wrong The comparison was dropped from the body on 2026-08-18 but survived in the key points of the Japanese version.
Removed the source for the 2-4 week figure and repointed the Corporation Tax Act link.
Why we got it wrong The document cited was the text of the General Incorporated Associations Act, which says nothing about how long the procedure takes, and the source was named as the Ministry of Justice while the link went to e-Gov. The link given for the Corporation Tax Act also pointed at its enforcement order, and now points at the Act itself (Act No. 34 of 1965).
The social-expenditure total was still the old figure in the closing section.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Clarified that the consumption-tax new-corporation notification does not apply to a general incorporated association
Why we got it wrong The rule in Article 12-2 of the Consumption Tax Act turns on stated capital or contributed capital of 10 million yen or more. A general incorporated association has neither, and its endowment fund is neither. Treating the fund as capital risked prompting readers to file a notification they do not owe.
The culture budget and the international shelter comparison were still the old figures in the social copy.
Why we got it wrong The Swiss and Finnish figures were dropped from the body on 2026-08-18 but survived in the social copy.
The number of dual-income households was still the old figure in the Japanese social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Matched the wording of Article 45 of the Income Tax Act to the statute and added the source.
Why we got it wrong The article said it was quoting the statute but used a term the statute does not use. The wording now follows the text on e-Gov law search, which is cited. The statement about case law has also been changed, since absence was never confirmed.
The ownership rate, the number of expiring certificates and the trouble rate at medical facilities were still the old figures.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Corrected the plan filing window, the start of the payment application count, and the claim about paper-only filing, against MHLW primary sources. Added the conditions for split payments and the equipment subsidy.
Why we got it wrong Readers act on this guide with money at stake. Omitting the lower bound of the filing window made early filing look safe. The payment count starts the day after training ends, not on the end date. On electronic filing, the MHLW page the article cited does not contain the word "paper" at all, and states the opposite: "when filing electronically through the Employment-Related Subsidy Portal". The claim was written without opening the source.
Added a source for the 2010 child allowance amount.
Why we got it wrong The amount itself was right, but none of the article's fourteen sources carried it. The figure now cites the Ministry of Health, Labour and Welfare's own document of 17 November 2010, which states 13,000 yen a month per child through the end of junior high school, with no income cap.
Replaced the claim that Takagi Village "succeeded" with what was decided and when it applies, and corrected the village name.
Why we got it wrong The village's own notice records that the change takes effect after the June 2027 election. This article was published in April 2026, so "success" could not be claimed. The village name was also misromanised as "Takatogi". The sentence about municipalities in Hokkaido amending bylaws for remote attendance appears in none of the article's three sources and no individual case could be confirmed; it has been replaced with the February 2024 revision of the model rules of procedure by the national associations of council chairs.
The conflation of "visiting foreign nationals" with "all foreign nationals" survived in the summaries and social copy, and the suspended-sentence diagram still disagreed with the body.
Why we got it wrong The 43,622 figure covers penal-code offences by all foreign nationals; the count for visiting foreign nationals is 33,037. That distinction was written into the body on 2026-08-18 but not into ogLead, the key points or the social copy. The diagram was left at 89.4% and 80.8% after the body moved to 86.1% and 78.0%. The English body also still carried the "about 60%" share with a source attached; that share is a calculation the source does not publish, so it has been replaced with the per-nationality counts it does carry.
Sweden's textbook budget was still the old figure in the key points and social copy, and in a currency the source never used.
Why we got it wrong The body was corrected on 2026-08-18, but the key points and social copy were left on the old figure. They also gave it in euros, while the source (VOA/AP) states dollars and kronor. The Japanese and English versions had been carrying different amounts from the same source.
The old way of counting job seekers was still in the Japanese social copy.
Why we got it wrong The body of both versions was corrected on 2026-08-18, but the Japanese social copy still described the figure as an annual average. What is published is the monthly average of monthly active job seekers.
Brought the English figures into line with the Japanese, which had been corrected earlier.
Why we got it wrong The correction of 18 August 2026 stated that the mismatch between the Japanese and English figures had been resolved, but only the Japanese text was updated. The English version still carried the older population estimate and, more seriously, gave the FY2025 expenditure figure of 115.5415 trillion yen as if it were FY2026: the article's own image caption named it as FY2025. FY2026 general account expenditure is 122.3092 trillion yen. Tax revenue figures now cite the Ministry of Finance directly.
Corrected the Grade 3 figure to the published MEXT number.
Why we got it wrong MEXT's non-attendance figures by school year record 17,997 for Grade 3 in the 2023 school year. 19,492 matches no grade in any year. A source has been added to the table.
Added a source to the table by population band and marked the operating rate as this article's own calculation.
Why we got it wrong The Cabinet Office overview of adoption and operation as of end-March 2025 publishes total entities, adopted, adoption rate and operating counts. It does not publish an operating rate. All eight rows check out arithmetically, but readers cannot tell a published figure from a computed one when both sit in the same table. The note also shows how the figure changes if the denominator is the number that adopted.
Aligned the Microsoft 365 discount figure with Microsoft's own wording.
Why we got it wrong Microsoft states "up to 75 percent" and gives no floor. Writing "60-75%" reads as a guaranteed minimum of 60%. We also confirmed that from July 2025 the free grants for Business Premium and Office 365 E1 ended, leaving up to 300 granted Business Basic licences, which is what the table already said.
A comparison with the OECD average that had been dropped was still in the summaries and social copy.
Why we got it wrong The OECD-average comparison was dropped from the body on 2026-08-18 in favour of the JILPT index of manufacturing hourly wages at purchasing power parity. The old comparison survived in the summaries and the social copy.
Changed the denominator of the 85% and 15% from "the grant" to project cost (A).
Why we got it wrong JANPIA's Cost Estimation Manual for Fund Distribution Organizations states that "the grant application amount for grants to implementing organizations shall be at least 85% of the grant application amount for project cost (administrative expenses no more than 15%)." The denominator is project cost, not JANPIA's total grant (A + C + D). Writing "85% of the grant" reads as the wider base and yields the wrong figures in a budget.
The breakdown of reasons for school non-attendance was still the old figure in the Japanese social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The count of violation types and the cycling-infrastructure figures were still the old ones in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
A figure that had been dropped, an increase of about 200,000 yen a year, was still in the heading, summaries, figures, table and timeline.
Why we got it wrong The claim about a worker on 5 million yen paying about 200,000 more a year was dropped from the body on 2026-08-18 because the Toyo Keizai article cited is paywalled beyond the relevant passage. It survived in seven other places and has been replaced with the premium rates the sources do carry.
Stated in the summary and social copy which date the 938 citations were counted on.
Why we got it wrong The body said "as of March 2026", but the summary beside the title and the social copy gave 938 with no date, which reads as a current figure. Counting the same way in September 2026 gives 2,444. A note to that effect has been added to the body.
Corrected "mandates legal protection" of quiet areas to what the Directive actually requires.
Why we got it wrong Article 8(1)(b) reads "shall also aim to protect quiet areas against an increase in noise" — an aim that action plans must carry, not a protection mandate. The same article adds that "the measures within the plans are at the discretion of the competent authorities," and Article 3(l)(m) leaves both the delimitation of a quiet area and its threshold to the Member State. "Mandates legal protection" claimed more than the Directive provides.
The core-farmer count and average age were still the old ones in the title, the summaries and the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
A dropped commuter-cyclist count and an old figure for cycling infrastructure were still in the Japanese summaries and social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The Middle East dependence rate was still the old figure in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The university enrollment rate was still the old figure in the title, the summaries and the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Brought the wage figures into the units the source uses.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Replaced the dates of supposed ministry directives with what the primary document actually says.
Why we got it wrong What happened in 2014 was the amended Public Assistance Act taking effect, not a directive. For 2023, what can be confirmed is a restatement in the materials for the division chiefs' meeting, and the wording about keeping forms readily available does not appear in that document. The passage carried no source, so it has been rewritten to what the primary document supports and the document is now cited.
Dropped the "average NPO salary of 2.31 million yen" where it survived in the summaries, the social copy and one body paragraph.
Why we got it wrong The 2.31 million figure was dropped as unsupported on 2026-08-18, but only from one body paragraph; it survived in the key points, the social copy and the "lack of specialization" section. The English version was in the same state, and additionally still carried 49,257 certified NPOs and unsourced estimates for dissolutions (about 1,500 a year against about 1,200 new certifications). These have been replaced with the counts the Cabinet Office publishes.
The loneliness share was still the old figure in the key points and social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The billionaire wealth total was still the old figure in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
The non-regular employment share was still the old figure in the social copy.
Why we got it wrong The body was corrected on 2026-08-18, but ogLead, the key points and the social copy — the places a reader sees first — were left on the old figures.
Replaced all six source URLs and corrected the Hamamatsu asset scope and the Miyagi lead company against primary sources.
Why we got it wrong All six URLs returned HTTP 200, so the link-reachability check passed them. Opening and reading each one showed the mismatch: the URL cited as the "Hamamatsu City Sewerage Concession Project", for example, was a page about a report on comprehensive private outsourcing of sewer pipe maintenance. The figure "seven" was the number of service areas absorbed in the 2005 merger of twelve municipalities, not the number of facilities under the concession. The Miyagi lead company also contradicted the prefecture's own site.
Rewrote the description of which ministries promoted concessions, to match the actual division of responsibility.
Why we got it wrong The Ministry of Children and Families has no role in water supply administration. Water supply development and management was transferred to MLIT and the Ministry of the Environment (MLIT White Paper 2024, column on the transfer of water supply administration). As written, the sentence would have misled readers about which ministry is responsible.
The Susaki case was described as water supply; it is sewerage. The description of tariff revision in Miyagi was the reverse of what the prefecture publishes. The comparison tables were rebuilt.
Why we got it wrong Susaki's primary sources — the city's own site and the project selection and request for proposals — contradicted the article on the project name, the utility covered, the contract period, and the cost allocation. For Miyagi, the prefecture's official site states plainly that "the private company has no authority to revise tariffs", and the article said the opposite. Because the comparison of three cases is the spine of this article, the tables themselves were rebuilt from primary sources.
Updated the cumulative figures to the final numbers published by the Japan Trust Association, and moved the citation to the document that actually carries them.
Why we got it wrong The URL we cited was a product description page and did not carry the figures. The numbers appear in Appendix 3 of the association's quarterly Trust Business Overview, which we now cite. The scheme closed at the end of March 2026, so the March 2026 figures are the final cumulative totals. The ¥6.43 million average had no source and could not be derived from the totals quoted in the same sentence (the old figures divide to about ¥7.50 million); it is now written so that readers can see it is a calculation.
Moved the median-financial-assets citation to the page that carries the figure, and stated which aggregation it comes from.
Why we got it wrong The figure itself (¥7 million) is correct and was confirmed in the source spreadsheet. The same survey also reports a median of ¥12.7 million for households that do hold financial assets, so the aggregation has to be named or readers will take the wrong number.
Removed three "system changes" that do not appear in the 2026 public offering guidelines, and rewrote the article based on the guidelines.
Why we got it wrong This article was published on 1 April 2026, while the public offering for 2026 fund distribution organizations (regular category) opened on 3 April 2026. The article described "this year's changes" with specific figures before the guidelines were published, without supporting evidence. On checking the published guidelines, two of the three items have no corresponding description, and there is no statement that the application method changed in 2026.
Corrected the self-funding ratio to at least 20% of project costs, as stated in the guidelines.
Why we got it wrong The guidelines state that organizations "shall, as a principle, secure at least 20% of the total project cost over the grant period from their own funds or private funds." Our introductory guide already stated 20% or more, so the two articles contradicted each other.
Removed the statement that "the indirect cost rate is generally 15% in recent years."
Why we got it wrong The 2026 guidelines contain no category called "indirect costs." The closest provision is "administrative expenses" for fund distribution organizations, capped at 15% of the grant amount. The same figure had been applied to a different concept.
Replaced the adoption rate with the actual figures from published results.
Why we got it wrong The JANPIA Annual Report was cited as the source for "approximately 25-30%," but no such description could be confirmed. The figures have been replaced with published results. Because the ratio varies by round, the range has been removed.
Replaced the post-monitoring period with the asset disposal restriction stated in the guidelines.
Why we got it wrong No basis could be found for "one to three years." What the guidelines state is the obligation to manage acquired assets for five years after project completion, and for buildings, for the useful life of depreciable assets as defined in the Corporation Tax Act.
Refreshed the figures to the latest date the Ministry of Justice publishes.
Why we got it wrong The source page carries only the newest snapshot. When it moved to 31 July 2026, the 30 June figures the article cited disappeared from it. The grounds, land types and residual counts have all been moved to the same date.
Reframed the vacancy count and the prefecture tallies as our own calculation.
Why we got it wrong The 20,091 is the quota of 240,971 minus the 220,880 appointed; what the ministry publishes is the 91.7% fill rate. The prefecture counts are likewise ours, so both are now presented as the article's own arithmetic.
Replaced a share we had produced by adding three chart segments with the breakdown the source publishes.
Why we got it wrong The 41.5% was the sum of three segments of a stacked bar chart and appears nowhere in the source, which states only the 33.5% figure for FY2025. The article now gives the published shares. The decline and percentage for new members were likewise our own arithmetic and have been trimmed to the two published values. The joining and leaving series sit inside charts, so the pages were rendered and read by eye, with the values recorded in the registry.
Replaced the peak volunteer firefighter figure with the actual intake, departure and establishment numbers the source publishes.
Why we got it wrong Only the peak headcount was given, so the article's own subject — the gap between people joining and leaving — could not be followed in figures. The published counts are now shown.
Added the source's notes and detailed counts on fiscal-year appointed staff.
Why we got it wrong Only the figures by sex had been quoted, dropping the source's note that some records have no sex recorded. The note and the detailed breakdown have been added.
Replaced the account of Type A dismissals, wages and the fee revision with the figures the ministry publishes.
Why we got it wrong The figures describing the scale were not tied to any source. They have been replaced with the monthly counts and the actual wage and allowance levels published by the ministry.
Corrected the breakdown of firms below quota, the actual employment rates in public bodies, and the description of the exclusion-rate system.
Why we got it wrong The breakdown and the public-sector rates did not match the year or categories in the source. The exclusion-rate system was described as it used to be, so the passage now sets out its history and the current rates.
Corrected the number of schools with pupils needing Japanese language support and the progression-rate comparisons.
Why we got it wrong The national comparison figures used in the progression rates came from a different year than the source. The distribution of schools by size and the change over time have also been added.
Replaced the description of the club-activity reform, including the activity-hour limits, with what the current national plan actually sets out.
Why we got it wrong The activity-hour limits came from an earlier guideline and are not part of the current plan. The passage now follows what the cited plan states.
Replaced the bare multipliers for the growth in foster carers with the counts the source publishes.
Why we got it wrong The 1.4x and 1.7x multipliers had been attached to the wrong quantities. The counts are now given so that it is clear which figure grew by how much.
Updated the figures on abuse deaths of infants under one to the latest report and added the breakdown.
Why we got it wrong The 504 deaths were the cumulative total to the twentieth report; a further tranche has since been added. The denominators for the percentages also did not match the report's categories, so the passage now follows those categories.
Added the case counts the source gives alongside the percentages for guardianship applications and appointments, and corrected the hearing-length figure.
Why we got it wrong The article gave percentages without their denominators. The source publishes both counts and percentages, so the counts have been restored. The 37.9% hearing-length figure did not match the source's categories.
Checked every care manager exam figure against the ministry's table row by row and added the years and cumulative totals that were missing.
Why we got it wrong Every figure in the article matched the source table. The article covered only the most recent years, however, omitting the lowest year and the cumulative totals, which have now been added from the full table.
Replaced the minus 0.0% year-on-year change in fraud arrests with the actual count, and added the breakdown of losses.
Why we got it wrong Minus 0.0% was a single-case decline expressed as a rate, which carries no meaning as a rate. It has been replaced with the count the source gives.
Updated the Fire and Disaster Management Agency casualty figures to the latest bulletin.
Why we got it wrong The bulletin is revised as new reports come in, and the figures the article quoted from the 35th report were out of date.
Confirmed the unit correspondence for the payment-month figures.
Why we got it wrong The 71.28, 89.34 and 96.52 million figures restate the ministry's 7,128, 8,934 and 9,652 expressed in units of 10,000 months. The one-to-one correspondence has been checked and recorded in the registry.
Aligned the liquidation asset figure with the source's wording and added the procedural history.
Why we got it wrong The source says just over 83.4 billion yen, which the article had rounded down to about 83.4 billion. The procedural history has also been added.
Corrected the evacuation rates for each earthquake to the figures each survey publishes.
Why we got it wrong Evacuation rates are defined differently in each survey depending on when the person left, yet they were presented as the same measure. For the 2003 Tokachi-oki earthquake only the highest municipality was quoted; the lowest is now shown alongside it.
Added the source's qualification and the agricultural damage figures to the account of the Tanna Tunnel works.
Why we got it wrong The source hedges these figures as reported estimates, so stating them as established values was too strong. The scale of agricultural damage has also been added.
Replaced the summed total for livestock veterinarians with the per-category counts the source publishes, and removed the prefectural percentages.
Why we got it wrong The 7,771 figure was the sum of two categories, a total the source does not give. The prefectural shares and the 3.3% and 5.9% declines were also our own calculations.
Corrected the number of bears culled under permit from 14,745 to 14,741 and added the monthly and prefectural breakdown.
Why we got it wrong Two places in the same article carried different figures; the source gives 14,741.
Restated the number of teachers leaving because of mental illness with the breakdown by school type.
Why we got it wrong The 1,319 and 950 figures were sums across school types that the source does not publish as totals, and the increase of 369 and the 8.5% share were derived from them.
Corrected fiscal years to calendar years, as the source chart labels them, and added the required headcount.
Why we got it wrong The Japan Bus Association chart runs 2017 to 2030 on a calendar-year axis, not fiscal years. The figures themselves (116 / 93 / 36 thousand) match the chart. The required headcount of 129,000 has been added so the shortfall has a stated basis.
Restated the Kumamoto travel-mode shares using all three survey rounds the source publishes.
Why we got it wrong The article skipped the middle round and connected only the first and last. Public transport was flat in the middle round, so the trend reads differently once all three are shown.
Replaced a figure we had produced by adding three chart slices with the breakdown the source actually publishes.
Why we got it wrong The 73.2% was the sum of three slices of a pie chart (20.4 + 26.4 + 26.4) and appears nowhere in the source, which states only that more than seven in ten were built in 1980 or earlier. The article now gives the three published shares. The 73.2 that does appear in the full report is the share of wooden buildings among vacant homes built in 2011 or later — a different table.
Corrected the statistical category name for vacant homes, the vacant-home bank participation figures, and the cumulative transaction count.
Why we got it wrong The category name did not match the official one used in the statistics. The 3.3% and 38.7% figures and the vacant-home bank counts could not be confirmed, and have been replaced with survey results that can be.
Replaced the 61.8% uncertain share with the surveyor's own answer options, repointed Yokohama's outflow at the ministry survey, and removed Kawasaki's figure.
Why we got it wrong The 61.8% was the sum of three answer options; the surveyor writes only 'about 60%'. The article cited for the Yokohama and Kawasaki figures is behind a content block and its text could not be retrieved. Yokohama's figure was confirmed in the ministry survey and repointed there; Kawasaki's 13.5 billion yen could not be confirmed and has been dropped. The Kanagawa total is now given exactly as the source states it.
Moved three logistics and construction figures to documents that carry them, and matched the published values.
Why we got it wrong All three were cited to index or press-listing pages carrying no figures. They now point at the Revised Logistics Efficiency Act pamphlet, the labour unit price press release, and the redelivery rate release. The 125-hour and 44% figures are both FY2028 targets, now stated as such. The '22.9% since 2021' was a compounding of annual growth rates that appears nowhere in the source, replaced with the published weighted average and year-on-year change.
Removed three of the four substance counts for labelling and SDS obligations and replaced them with figures that can be verified.
Why we got it wrong The 896 matched the file size marker [896KB] on the ministry's circulars page. The 1,600 was a chemical registry number in a different PDF, and the 2,900 appears in no source. Cumulative totals derived by addition have been dropped in favour of the figures the sources actually publish.
Corrected per-capita prefectural income, average pay by industry, and the settlement rates and career paths of regional revitalisation volunteers.
Why we got it wrong The income and volunteer figures were out of date. The 2.4-fold and 3-fold gaps were our own divisions and do not appear in the sources.
Corrected the number of fisheries cooperatives, the number of specified skilled workers, world fisheries output, and the workforce in the Iwami area.
Why we got it wrong The cooperative and workforce figures came from mismatched years. The claim of a 45% fall over twenty years was a calculation absent from the source, with no stated baseline.
Moved two citations to documents that carry the figures, and corrected which OECD average the 41-hour figure refers to.
Why we got it wrong The enrolment figures were cited to a page explaining the special-needs system, which carries no numbers; they now point at MEXT's Special Needs Education Data compiled from the School Basic Survey, with FY2024 added. On TALIS, 41.0 hours is the average across participating countries for elementary teachers; the OECD average for junior-high teachers is 40.4 hours, making the gap 14.7 hours. The OECD page refuses requests, so the citation now points at the Japanese national report on the same survey.
Corrected the currency conversion of Seoul's loneliness budget and the Japanese platform subsidy figures to what the sources state.
Why we got it wrong The dollar and yen conversions were ours; the source gives only the dollar figure as reported by CNN. The Japanese subsidy is published as a per-authority ceiling, not a national total.
Replaced the 15-20% take-up rate for public assistance with the indicator name and ranges the source uses.
Why we got it wrong The source does not use a take-up rate; it reports a household coverage ratio, separately for an income-only test and a test that also counts assets. The value moves a great deal between the two, but only one range was given.
Corrected the private afterschool fee bands, the dual-earner household figures, and the fall in income after childbirth.
Why we got it wrong The fee bands came from the paywalled part of the article and could not be verified, so the passage now keeps to the headline and lead. The dual-earner figures were cited to a chart showing dual-earner versus homemaker households, which carries no breakdown by working hours. The report states a fall of roughly 60% after childbirth; the figure of 67.8% appears nowhere in it.
Reframed the top-three share and the fine total as our own calculation.
Why we got it wrong The 87% is the sum of three published shares and the ¥14.58 million is counts multiplied by fine levels; neither is published by the police. Both are now presented as the article's own arithmetic rather than as source figures.
Corrected the family-policy spending shares of GDP for France and Japan to the year and definition the source uses.
Why we got it wrong Neither 2.9% nor 1.6% matched any year or definition in the source. Family-policy spending varies with the definition used, so the figures are now given per definition.
Corrected the household share of higher education costs, teacher starting salaries, class size, and the share of non-teaching duties.
Why we got it wrong The source publishes teacher pay as a range from minimum to maximum, but the article compared only the minimums and called the gap large. The class size is 27, and the 71% figure for non-teaching duties does not appear in the source.
Updated total social security benefits and their funding breakdown to the latest published year.
Why we got it wrong The published figures have been revised, and the share of GDP has moved with the revision of the denominator.
Restated the digital-services trade deficit using the figures and range the source gives.
Why we got it wrong The figure was written to the last hundred million yen even though the total depends on which service categories are included. It is now given as the source presents it.
Corrected the year of the related-population estimate from 2024 to the FY2023 survey.
Why we got it wrong The same figure comes from the FY2023 survey. The year has been aligned with the source.
Corrected classroom Wi-Fi coverage from 95.4% to 96.2%.
Why we got it wrong The 95.4% belonged to a different line on the same chart (wired in-school LAN) and to an earlier year. The wireless figure has been taken from the correct year.
Corrected the levy rates for FY2027 onward and the growth rate of earned income to what the sources state.
Why we got it wrong The 0.37% and 0.44% rates appear neither on the Children and Families Agency page nor in the MHLW explanatory materials; both give the FY2028 contribution of ¥900 billion and illustrative amounts by income. The published illustration for a ¥6 million income in FY2028 is ¥1,000 a month, not ¥1,100. Earned income grew at 0.4% a year over 24 years, not 0.5% over 25.
Corrected the outflow amounts and cost breakdown for the hometown tax scheme to the published figures.
Why we got it wrong The Yokohama and Tokyo amounts were from the wrong year. The cost breakdown percentages also did not match the denominator the source uses.
Corrected the counts for long absences, students with no specialist consultation, and ICT-based attendance recognition, and moved the citation from a notice page to the survey summary.
Why we got it wrong The figure of about 67,000 students absent 90 days or more conflated the count with high-school non-attendance (67,782); the published figure is 191,958, or 54.2%. Students receiving no specialist consultation number 135,724, and the 38.3% share is printed in the source. The 10,409 ICT attendance figure was FY2022; the FY2024 figure is 13,261. The 3.8% enrolment share and 'one in 26' appear nowhere in the source and have been replaced with the published 38.6 per 1,000. The cited URL was a notice page carrying no figures and now points at the survey summary.
Corrected the shares of teachers able to use ICT and those who had taken training, to match the survey year cited.
Why we got it wrong The survey year cited and the year the figures came from did not match. The prefectural high and low figures have been taken again from the correct year.
Corrected the cumulative education-fund trust figures and the amounts in the 2022 Supreme Court tower-condominium case to the itemised figures in the source.
Why we got it wrong The three case figures were all totals, and the cited site could not be retrieved. They have been replaced with a scholarly article that recites the judgment's findings, with the components sourced individually. The assessed-road-price valuation is in fact 333,706,241 yen; the article had rounded off the last three digits. The trust totals were also out of date.
Corrected the share of care providers struggling to recruit, the reasons given, the pay gap with all industries, and the number of foreign care workers.
Why we got it wrong None of the percentages matched the survey question or year in the source. The pay gap is now given as the two actual figures the source publishes rather than as a derived difference, and the foreign worker count has been set to the published number.
Corrected Okinawa's real annual average from about 1,005 yen, and noted that the 181-day gap is our own count rather than a figure from the source.
Why we got it wrong The 1,005 yen figure is Ibaraki's pre-revision minimum wage as listed on the source page. The source calculates only Akita's 991 yen and gives no real annual average for Okinawa, so the 14-yen gap was derived from a figure that does not apply. Applying the same formula to Akita reproduces the source's 991 yen, and the Okinawa figure is now labelled as our own calculation.
Corrected the concentration of foreign overnight stays in the three largest metropolitan areas, accommodation tax revenue, overseas city tax rates and revenue, and land price growth.
Why we got it wrong The concentration figure did not match the source's definition. The overseas revenue figures included yen conversions of our own, and the land price growth rates did not correspond to the survey points in the source.
Moved the poverty figure to the survey summary and replaced the ODA total with JICA's published breakdown.
Why we got it wrong The poverty rate was cited to an index page carrying no figures. The ¥3.9 trillion total does not appear in the JICA release cited, so the breakdown that release does publish is used instead. Figures cited to MOFA and OECD pages remain unverified because both refuse automated requests.
Corrected the vacancy rate to the prefecture figure the source publishes.
Why we got it wrong The 14.1% belongs to Nara in the national results cited; Kyoto Prefecture stands at 13.1%. The source carries no city-level breakdown, so the figures for Kyoto City alone have been dropped.
Moved the headcount citation to a document that carries the figures and dropped resignation counts from a mismatched year.
Why we got it wrong The ministry survey cited is the FY2021 edition, while the article's figures are for FY2022; the FY2022 edition could not be located, so the counts have been dropped. The headcount page renders its tables in JavaScript, so the citation now points at the FY2025 staffing survey key-points PDF, whose chart and table were rendered and read by eye for the 3,282 thousand peak in 1994 and the 2,809,298 as of April 2025. The OECD general-government employment share remains unverified because the site refuses automated requests.
Corrected the competition ratio for local government exams, staff numbers, starting salaries, and the number on long-term sick leave.
Why we got it wrong The starting year and value for the competition ratio did not match the source. Staff numbers were rounded and have been replaced with the published counts. The starting-salary comparison was dropped because the public and private surveys use different definitions.
Replaced the year-on-year change for the 1966 hinoeuma birth count with the actual figures for the surrounding years.
Why we got it wrong The minus 25.4% was a calculation of ours that does not appear in the source. The neighbouring years' figures are now given instead.
Corrected the birth counts, total social security benefits, their generational split, and the unmarried rate among non-regular male workers.
Why we got it wrong The social security total used the wrong year and classification, and the 10-trillion-yen child-related category is not a grouping the source uses. The 75.6% and 30.7% unmarried rates could not be confirmed.
Corrected the NATO Europe and Canada average, SIPRI's figure for Japan's military spending, and the Agency for Cultural Affairs budget.
Why we got it wrong NATO's table of defence spending as a share of real GDP gives 2.27 for 2025; 2.3 was a rounding. SIPRI states that Japan's military expenditure reached 62.2 billion dollars in 2025, so 62 billion was a truncation. The culture budget was out by one fiscal year: the FY2025 figure is 106.3 billion yen.
Updated the 2024 birth and marriage counts to the final figures, and corrected the child-rearing cost and family-policy spending as a share of GDP.
Why we got it wrong The birth and marriage counts were preliminary and have since been finalised. The GDP shares for family policy did not match the year or definition used by the source.
Corrected the child-rearing support levy rate and per-person amounts, France's family spending share, and Japan's birth count.
Why we got it wrong The levy rate did not match the source, and the burden is set as a monthly amount per scheme rather than a single rate. The claim of 110,000 social media posts could not be confirmed. The birth count has since been finalised.
Moved the minimum wage to MHLW's own announcement and dropped an international ranking that could not be checked.
Why we got it wrong The minimum wage was cited to a payroll vendor's article; MHLW announces the ¥1,121 national weighted average and the ¥66 increase itself. Japan's rank in the OECD average annual wage series could not be checked: the article cited runs to a second page that could not be retrieved, and the OECD site refuses automated requests. MHLW's white paper comparison on a 1996 = 100 basis is used instead.
Corrected the US estate tax exemption and removed the count of business succession plan applications.
Why we got it wrong The 11.18 million dollar figure was a conversion whose exchange rate and year could not be pinned down. The 21,748 applications could not be traced to a source and have been dropped.
Repointed the 106.3 billion yen culture budget at a document whose contents can actually be read.
Why we got it wrong When this figure was corrected earlier today, the PDF cited as its source turned out to be malformed and could not be read. It now points to the FY2026 overview, which states the figure for both FY2025 and FY2026.
Corrected the education budget, shelter counts and population coverage, the number of public libraries, and the New York Public Library budget cut.
Why we got it wrong The education budget confused the category that includes science and technology promotion with the one that does not. The shelter counts conflated all designated evacuation facilities with underground shelters. The coverage figures for other countries also did not match the sources.
Corrected the number of dual-income households and replaced the average monthly childcare fees with the income-based bands the source sets out.
Why we got it wrong No national average monthly childcare fee is published. What the state sets is an eight-band ceiling by household income, which cannot be written as an average.
Replaced a computed multiple with the published counts and updated the recruitment ratio to the latest published figures.
Why we got it wrong The 1.85 times was 3,827 divided by 2,065 and appears nowhere in the source; the counts and shortage rates are now given directly. The recruitment ratio was cited to an index page carrying no figures and now points at the latest implementation report.
Corrected the My Number card holding rate, internet use among older people, the number of cards due for renewal, and the share of clinics reporting problems.
Why we got it wrong The holding and usage rates were out of date. The 87.5% figure was derived by inverting the share of clinics reporting no problems, which is what the source actually publishes. The estimate of 16 million non-users, or 13% of the population, could not be confirmed.
Corrected the average pay for care workers and for all industries, and the share reporting physical strain.
Why we got it wrong The 386,000 yen all-industry average did not match the indicator in the source, and the roughly 80,000 yen gap was derived from it. The share reporting physical strain is 29.3% in the source.
Corrected the share of women in management and the female labour participation rate to the source's definition and year.
Why we got it wrong The definition of management did not match the source. What the source publishes is an employment rate for a different age band, not a labour force participation rate.
Updated the total length of water mains, the ageing rate, and the length and rate of replacement to the latest published figures.
Why we got it wrong All of these had been revised since publication. The pace of replacement needed has also been added for comparison.
Updated the 2025 spring wage round figures from the first tally to the final one.
Why we got it wrong All three figures came from the first tally in March 2025; the final tally published on 3 July 2025 gives different values. The claim of the highest level in 33 years does not appear in the source and has been replaced with a comparison to the previous year. The size bracket is also under 300 employees in the source, not 299 or fewer.
Corrected the regional consumer price indices and the electricity bill comparison.
Why we got it wrong The lowest index belongs to Gunma, not Kagoshima City. The electricity figures were calculated from contract assumptions of our own and differed from the published amounts.
Corrected the pension reserve's assets under management and cumulative returns, long-term care benefits, and the generational accounting figures.
Why we got it wrong The asset and return figures were out of date, and the generational accounting amounts did not match the cohorts or values in the source.
Corrected the household share of education costs, the total education expenditure, the deprivation rates for single-parent households, and the drop in mothers' earnings after childbirth.
Why we got it wrong The education totals were rounded approximations that did not match the published amounts. The deprivation rates did not correspond to the survey questions, and the earnings-drop figures did not match the source's categories.
Dropped the 25.9% uncontested-election rate and replaced it with the district and seat counts the source publishes.
Why we got it wrong The 25.9% figure does not appear in the source. What the source gives is the number of uncontested districts and the number of members elected without a vote.
Dropped the share of general account revenue attributed to consumption tax.
Why we got it wrong The 33.3% does not appear in the finance ministry document cited, which tracks the national burden ratio. The 5.7% and 2.1% from the household survey also remain unverified, as the statistics bureau page is a link list rather than a table.
Corrected the arrest counts for crimes by foreign nationals, the Japanese arrest total, the nationality breakdown, and the suspended-sentence rate.
Why we got it wrong The 43,622 figure covers crimes by all foreign nationals; restricted to visiting foreign nationals it is 33,037. The categories had been conflated. The claim that two countries account for about 60% was a calculation absent from the source, which instead publishes counts by nationality.
Replaced two annual-salary figures and the 600-trillion-yen retained-earnings figure with the monthly wages and ordinary profit the sources actually publish; also corrected the source for the 2.3% nominal wage rise.
Why we got it wrong Neither survey publishes annual salaries. The 4.05 million yen figure was the monthly 338,200 yen multiplied by twelve; the 6.60 million yen figure appears in no source. The two also measure different things (scheduled pay versus pay including bonuses), so they cannot be compared as annual salaries. The 600 trillion yen came from a page listing research papers, with no such figure on it. The 2.3% rise was taken from a single-month December release and presented as the full-year figure.
Corrected the classroom Wi-Fi coverage rate, Sweden's textbook budget, and the share of pupils reporting fatigue after digital-textbook lessons.
Why we got it wrong The 95.4% belonged to a different line on the same chart (wired in-school LAN) and to March 2021, not March 2023; the wireless figure for March 2023 is 95.7%. The textbook budget differed between the Japanese and English versions while citing the same source, and that source cannot be retrieved. The fatigue figures took only the upper end of ranges the ministry report states as 20-30% and 20-40%.
Restated the job-seeker count to match the published table.
Why we got it wrong The Japanese version said about 1.93 million and the English 1.76 million. Both now follow the published monthly averages of 1,929,824 for 2024 and 1,906,836 for 2025.
Corrected the total population to the published figure and moved the citation to the monthly report.
Why we got it wrong The population estimates page renders its tables in JavaScript. The citation now points at the monthly PDF and follows the confirmed figure. The Japanese and English versions had also carried different numbers (123.22 million against 122.6 million); both now match the source.
Corrected the counts of municipalities at risk of disappearing and the projected fall in the young female population.
Why we got it wrong The figure of about 282 municipalities was the sum of 165 and 117, a total the source does not give. The 87.5% was stated without saying which municipality or what population it referred to; the numerator and denominator are now shown.
Replaced the fertility rate description with what the source states.
Why we got it wrong The 1.26 is an annual vital statistics figure and does not appear in the municipal report cited; the text now follows that report's distribution and its two highest municipalities. Tokyo's 0.99 remains unverified because Nikkei is behind a paywall.
Stopped presenting the 1990 premium and the founding rate as source figures.
Why we got it wrong Neither the ¥36,150 nor the 3.4% appears in the Cabinet Office tax commission material cited. The former is a calculation from the rates then in force and the latter a widely known figure, so neither is now attributed to that source.
Replaced the 0.4% fraud share of public assistance spending with the counts and amounts the source publishes.
Why we got it wrong The 0.4% does not appear in the source; it was the amount divided by the total. The published counts and amounts are now given alongside the total.
Updated the education cost totals to the revised figures and replaced the OECD wage comparison, whose source site has closed, with JILPT's.
Why we got it wrong MEXT revised these education cost figures on 16 January 2026; the 5.96 million, 19.76 million and 1.03 million yen figures were on the pre-revision basis. The wage comparison cited an OECD statistics site that has since closed and could not be verified, so it has been replaced with JILPT's index of hourly manufacturing wages on a purchasing power parity basis.
Moved the tuition caps to a document that carries them and dropped beneficiary counts that could not be found.
Why we got it wrong The caps were cited to a system landing page carrying no amounts; they now point at MEXT's leaflet, with admission fees added. The figures of 340,000 rising to 750,000 beneficiaries appear neither on the JASSO page nor on MEXT's system page. Figures cited to the OECD country note remain unverified because the site refuses automated requests.
Corrected the GDP-basis burden rate, the national income basis rate, Japan's OECD ranking, the initial pension contribution rate, and the 13.5 trillion yen in transfers from the working-age population.
Why we got it wrong The 34.5% was Costa Rica's national-income-basis figure in the ministry's international comparison chart, where the bar label is the national income basis and the bracketed figure is the GDP basis. The 46.8% was a projection at the time; the later actual figure is 45.7%. The 4.9% initial pension rate appears nowhere in the rate table, whose first row is 64.00 per mille, or 6.4%. The 13.5 trillion yen figure does not appear in the cited report at all.
Corrected the suicide rate and the count of school-age suicides to MHLW's published figures.
Why we got it wrong The rate of 15.4 per 100,000 differs from the published 16.4 for 2024, and school-age suicides numbered 529, not 527. The OECD comparison has been dropped because the WHO statistics site refuses automated requests; the two cited papers also remain unverified because PubMed returns a challenge page.
Moved the pre-departure fee and disappearance figures to the surveys that actually publish them.
Why we got it wrong The expert council's final report, cited for the ¥540,000 average, contains no monetary figure. The agency's own survey of costs paid by trainees gives ¥542,311 and a breakdown by nationality. The disappearance count was cited to an MHLW page on labour-standards inspections, which does not cover disappearances; it now points at the Immigration Services Agency statistics, with the post-2023 decline added. The claim of more than 100,000 since inception could not be verified and has been dropped.
Updated the number of foreign workers, technical intern trainees, and specified skilled workers to the latest published figures.
Why we got it wrong All three had been updated since publication. The article gave rounded approximations, which have been replaced with the published counts.
Moved the 2022 disappearance figure to the statistics that publish it.
Why we got it wrong The expert council page cited is a redirect stub returning 42 characters. The figure now points at the Immigration Services Agency's published statistics, with the rate against resident trainees added.
Corrected the reasons given for school non-attendance to match the survey's current questions and options.
Why we got it wrong The 51.8% came from the survey's earlier answer options. The options have since changed, and the equivalent figure under the current categories is 32.2%.
Removed the count of offence types and the road-length figures, and replaced the fine amounts with the schedule set out in the ordinance.
Why we got it wrong Neither the figure of about 113 offence types nor the road lengths (78,638 km, 475 km, 0.6%) could be confirmed in any source. The fine amounts are set out in the ordinance and have been taken from there.
Aligned the bottom-quintile figure with the 'about twice' wording the source uses.
Why we got it wrong The ¥59,000 is half of ¥118,000 and appears nowhere in the Daiwa report, which states only that the top quintile receives about twice the bottom quintile. The citation now points at the PDF.
Moved the flat-rate tax cut to a document that carries the amounts and dropped an unverifiable premium increase.
Why we got it wrong The tax cut was cited to a landing page carrying no amounts; it now points at the National Tax Agency leaflet. The claim that a worker on ¥5 million paid about ¥200,000 more a year could not be checked: the Toyo Keizai article cited is an interview and is paywalled beyond its opening.
Updated the core-farmer and abandoned-farmland figures to MAFF's latest published statistics.
Why we got it wrong The 1.02 million workers, average age of 69.2 and 71.7% aged 65 or over match no year in MAFF's published series; the article now follows the ministry's basic data table. Both the structural survey page and the abandoned-farmland page cited were index pages carrying no figures. The farmland bank consolidation rate remains unchecked because that particular MAFF page refuses automated requests.
Removed the annual premium amounts from the payslip table and replaced them with the contribution rates the sources publish.
Why we got it wrong The annual amounts were calculated from an assumed salary of 5 million yen and do not appear in any source. What the sources publish are the rates.
Corrected the cultural budget to the published figures and dropped the Louvre's percentage increase.
Why we got it wrong The ¥110 billion differs from the published figures and does not appear in the PDF cited; the text now follows the FY2026 budget overview. The Louvre's 38% increase could not be confirmed because the museum's site renders its prices in JavaScript.
Moved the consumer price citation to a document that carries the figures.
Why we got it wrong The citation was an index page carrying no figures; it now points at the national annual average PDF, with the overall rate added. The Teikoku Databank count of over 20,000 items remains unverified.
Split the figure on out-of-hours contact into the overall rate and the rate for men.
Why we got it wrong The figure for men had been presented as the overall figure. The source publishes both.
Corrected the deflator paired with the real-wage figure, dropped two figures absent from their sources, and re-attributed the productivity-to-wage sensitivity to MHLW rather than the OECD.
Why we got it wrong The 1.3% real-wage decline is deflated by the consumer price index excluding imputed rent, which rose 3.7%; the 3.2% the article paired with it belongs to the all-items index, which yields a 0.8% decline. Two different indices had been combined. The OECD comparison (+38.7% for South Korea and so on) and the service-sector pass-through rates (about 30% against nearly 100% in the US) appear nowhere in the cited sources and could not be confirmed in the publishers' material, so both have been dropped; the international comparison is replaced with the one MHLW's white paper publishes on a 1996 = 100 basis. The statutory non-wage labour cost series has been dropped: the Keidanren page cited reports year-end bonus settlements, and the survey it would have come from ended in 2019. The 0.4% wage response to a 1% productivity gain comes from MHLW's white paper, not the OECD. The 5.25% wage settlement matches Rengo's summary, but the '33 years' framing describes the previous year, and the sentence has been rewritten accordingly.
Reported by Rengo 2025 Spring Struggle Summary; MHLW Monthly Labour Survey (preliminary, 2025); MHLW Analysis of the Labour Economy 2023
Added the correlation coefficients the source gives for productivity and real wages.
Why we got it wrong The link was described only qualitatively, but the source states it as a correlation coefficient, which is now given.
Stopped presenting the ¥170 subsidy target as a figure from the price survey.
Why we got it wrong The ¥170 is the government's target for holding retail prices down, not a price published by the Agency for Natural Resources and Energy's survey. The survey is no longer cited for it.
Removed five figures — bicycle commuters, cycling lane length, housing on narrow streets, cycling among over-60s, and Dutch cycle-path length — that could not be verified.
Why we got it wrong None of the five could be found in the sources cited, and no replacement source was found, so the figures have been dropped rather than re-attributed. Only the 7,570 km of cycling space, which is confirmed, remains.
Replaced the health insurance enrolment figures with the per-scheme counts the source publishes, and pointed the levy and single-person household figures at the underlying documents.
Why we got it wrong The 54 million figure for employment-based insurance is 77.36 million when the source's own rows are added up, a gap of more than 20 million. The source publishes enrolment by scheme, so the schemes are now listed separately. The 1 trillion yen levy and the single-person household count were correct; both had been cited to landing pages and now point to the documents themselves.
Added the neighbouring age bands to the My Number health card usage rate.
Why we got it wrong Only the lowest band had been quoted. The source publishes the rate for each band, so the full gradient is now shown.
Moved the gender pay gap citation to a document that carries the figure.
Why we got it wrong The citation was an index page carrying no figures; it now points at the summary PDF, with the wage levels for men and women added.
Recorded that the referendum result and construction progress could not be matched to their sources.
Why we got it wrong Neither the 71.7% opposition figure nor the 6.6% construction progress appears in the retrieved text of the nippon.com and Akahata articles cited. The figures are widely reported, but the reason they remain unverified here is recorded in scripts/unverifiable-statrefs.json.
Corrected the population decline in affected municipalities, the population of the Oku-Noto area, and the number of evacuees.
Why we got it wrong The claim that nine in ten affected municipalities lost population was conflated with the national trend. The Oku-Noto figure came from a paywalled article where only the headline was readable, and has been replaced with a report carrying the same content. The evacuee count has been updated to the latest published figure.
Corrected the volume through the Strait of Hormuz, Japan's Middle East dependence, and its oil stockpile, and removed the GDP estimate.
Why we got it wrong The volume and stockpile figures came from mismatched years, and the dependence rate is 94.7% or 95.2% depending on the year. The GDP estimate could not be confirmed and has been dropped.
Added the cut-off time and the channels used for the roughly 200 complaints received by the city.
Why we got it wrong The three-day window does not appear in the source, which instead gives a cut-off time and the channels through which the complaints arrived.
Corrected how the food self-sufficiency rate, the average age of core farmers, and food waste are presented.
Why we got it wrong The self-sufficiency rate varies widely with the basis used (38%, 46%, or 64%), yet only one figure was given. The article now shows the figures for each basis as the source presents them.
Updated the university and tertiary enrolment rates to the latest published figures, removed the claim that the rate was a record, and corrected the year label on the turnover rate.
Why we got it wrong MEXT revised the formula for calculating enrolment rates in its FY2025 release and recalculated earlier years. The 57.7% and 83.8% figures were on the old basis. On the new basis the rates are flat and falling respectively, so they cannot be described as a record. The 34.9% turnover figure is correct but tracks the cohort that graduated in March 2021 over three years, so the year label has been fixed.
Removed the claim that 75% of parents in non-participating municipalities feared a drop in quality, and replaced it with the published data on free school meals.
Why we got it wrong The survey cited could not be located and the origin of the 75% figure could not be established, so it has been dropped in favour of figures that can be verified.
Moved the wage figures from an index page to the employment-type section and restored the published units.
Why we got it wrong The citation was an index page carrying no figures; it now points at the employment-type PDF. The ¥349,000 and ¥233,000 were roundings of the published ¥348.6 thousand and ¥233.1 thousand, now stated as published.
Replaced the renewable share and FIT surcharge with published figures, and removed the 15% curtailment figure for Kyushu.
Why we got it wrong The 22.9% and 2.7 trillion yen figures were cited to the Energy White Paper's contents page, which carries no figures in its 1,293 characters. They have been replaced with the Agency for Natural Resources and Energy's subcommittee material and METI's surcharge announcement. The 15% curtailment figure does not appear on the Kyushu Electric page cited and could not be confirmed in METI materials, so it has been dropped.
Corrected the net inflow to Greater Tokyo and the population projection figures.
Why we got it wrong The 63 million figure was cited to a landing page with no figures on it, and the check had passed only because a candidate derived from 100 million matched the postcode in the page footer. The figure itself was right, so each value has been sourced to the document that carries it. The net inflow has been set to the published count.
Replaced the rounded per-capita medical costs by age with the published amounts.
Why we got it wrong Both figures were rounded so coarsely that they were more than 40,000 yen away from the published amounts.
Moved the non-regular employment count to a document that carries the figure.
Why we got it wrong The page cited does not carry the 2024 average of 21.26 million; the citation now points at the summary PDF of the detailed tabulation, whose trend table does.
Corrected the shares for young people's regional mobility.
Why we got it wrong The end-point figures were each about half a point off the source, and have been corrected.
Corrected the household count to the published figure and moved the citation to the monthly release.
Why we got it wrong The 1.67 million differs from the published 1.65 million, and the PDF cited was an unrelated document on meal provision. The citation now points at the March 2024 provisional release. The 15–20% capture rate remains unverified because the repository page cited holds only bibliographic data.
Replaced the 0.5% fraud rate for public assistance with the counts and amounts the source actually publishes.
Why we got it wrong The cited government material publishes the number of fraud cases and the amount involved, but not a rate. The article stated a rate without a denominator, so it has been replaced with the figures the source gives.
Updated total individual donations and the number of certified NPOs to the published figures, and removed the unsupported claim that NPO staff earn 2.31 million yen on average.
Why we got it wrong The donation total was cited to the publisher's research index page, which carries no figures; the published total is 2.0261 trillion yen. The figure of 49,257 certified NPOs appears in no year of the official series, which gives 49,478 at the end of March 2025. The average NPO salary was cited to a National Tax Agency survey of private-sector wages that does not cover NPOs, and no survey publishing a per-person NPO average could be found, so the figure has been dropped in favour of the staff counts and total salary payments the Cabinet Office survey does publish.
Replaced a loneliness share we had produced by adding three categories with the published breakdown, and moved the household projection to the press release.
Why we got it wrong The 39.3% was the sum of three Cabinet Office categories; the source states only 'about four in ten'. The article now gives the published shares. The household projection was cited to a table index page and now points at the press release, with the figures for women and the 23.30 million household count added.
Recorded that the user count and the commission rate could not be matched to their sources.
Why we got it wrong The 97 million users is cited to the white paper's landing page, which holds only chapter links, and the 20% commission to a JFTC explanatory page carrying no figures. Both reasons are recorded in scripts/unverifiable-statrefs.json.
Updated internet advertising spend to the latest year.
Why we got it wrong A more recent year has since been published, so both years are now shown.
Dropped a contract ceiling and a defence budget figure that could not be checked against their sources.
Why we got it wrong The $200 million ceiling does not appear in the Anthropic statement cited, and the Pentagon site refuses automated requests, so it has been dropped. The $1.8 billion AI budget and 40% increase were cited to the Pentagon's news index rather than any specific document. The UN General Assembly vote remains unchecked because the UN pages return a JavaScript challenge; the reason is recorded.
Replaced the billionaire wealth totals with what the source actually states.
Why we got it wrong The $18.3 trillion, 16% and 81% figures do not appear on the Oxfam page cited; the text now follows what that page says.
Corrected the non-regular employment share and moved the spring wage citation to a document that carries the figures.
Why we got it wrong The 37.2% belongs to a different table in the source (an employment rate for 2017); the published share for 2024 is 36.8%. The spring wage round was cited to Rengo's yearly index page, which carries no figures, and now points at the summary PDF.
Attributed the PMax video asset cap to its actual publisher.
Why we got it wrong The source name read as an official Google announcement, but the link pointed to an advertising agency's blog. Neither Google Ads Help "About your Google Ads account limits" nor the Performance Max asset group page states a video limit per asset group. The figure was not removed, since the report itself exists — it is now attributed to whoever published it, with a note that it cannot be verified in Google's own documentation. The June 2025 blog citation was also consolidated onto the URL that the older path redirects to.
Replaced the source for the verification timeline with Google's official help page (the figure itself is unchanged).
Why we got it wrong Google's own help page states that most requests are reviewed within 3–5 business days, yet the article cited a third-party explainer instead. Where primary documentation exists, it should be cited. The figure matched the official statement, so no numeric change was needed.
Replaced the Ad Grants policy source link with a page whose content can still be verified.
Why we got it wrong The previously cited page no longer exists as a standalone article and now redirects to Google's generic help index. Because it still returns HTTP 200, a link checker does not flag it — yet readers could not reach the cited material. The claim itself is unchanged.
Corrected the Goodstack review period from 3–14 business days to 3–5 business days.
Why we got it wrong The article conflated two separate steps: Goodstack's eligibility verification and Google for Nonprofits' own application review. Google's official help page states that most Goodstack requests are reviewed within 3–5 business days, while the Google for Nonprofits application review is documented as 2–14 business days.
Added a note on how the eligibility rules apply to social welfare corporations that operate licensed nurseries.
Why we got it wrong Google's help pages refer nurseries to Google for Education, yet nurseries are absent from the exclusion list in the eligibility requirements for Japan, and the Education eligibility criteria do not cover licensed nurseries either. The article did not address this discrepancy, leaving affected corporations unable to judge whether they could apply.
Corrected the paid-tier pricing to reflect annual-commitment rates and the per-edition discount levels.
Why we got it wrong The article did not distinguish annual-commitment pricing from monthly billing (monthly rates are $4.20 and $7.40). It also stated a flat 75% discount for every edition, whereas Google publishes 75% or more for Business Standard, 72% or more for Business Plus, and 70% or more for Enterprise. The Japanese edition of this article carried a further error: yen figures had been converted from USD at roughly 136 JPY rather than taken from Google's published yen pricing.
Corrected the compliance features, Enterprise storage, and free-tier user cap in the plan comparison diagram.
Why we got it wrong The feature set and storage limits were written without checking Google's official pricing page. Vault is included from Business Plus upward and DLP only on Enterprise. The user cap also failed to reflect Google's own distinction between "up to 300 users on new accounts" and "up to 2,000 users on request."
Rewrote the application timeline as two verifiable stages instead of a single unsourced figure.
Why we got it wrong The "7–10 business days" figure came from a third-party explainer and could not be corroborated in Google's own documentation. The process actually runs in two stages — the Google for Nonprofits application review and the separate Ad Grants activation — and only the former has a published figure (2–14 business days). The unverifiable combined estimate was replaced with what the official source states.
Added a new section on social welfare corporations that operate licensed nurseries.
Why we got it wrong The ineligibility list was based on only part of Google's official help content. A separate page states that schools, nurseries, academic institutions, and universities should refer to Google for Education, and the article did not address this discrepancy — leaving social welfare corporations that run nurseries unable to judge whether they could apply. Checking the referral destination showed that Google Workspace for Education defines qualifying institutions as primary schools, secondary schools, and degree-granting higher education institutions, which does not include licensed nurseries.
Replaced three source links that no longer resolve to the pages they cited.
Why we got it wrong Two of the cited help articles no longer exist as standalone pages and now redirect to Google's generic help index. Because they still return HTTP 200, a link checker does not flag them — yet readers could not reach the cited material. They were replaced with pages whose content can still be verified. The claim that Performance Max became "available to all Ad Grants accounts since January 2025" could not be corroborated in Google's published material, so the timing assertion was removed.
Fully updated the body, diagram, and statistics to the FY2025 data (released July 15, 2026).
Why we got it wrong A provisional structural analysis based on FY2024 data was fully rewritten to match the latest published (FY2025) data. The FY2024 approved-claim count was revised to 1,056.
Reflected the FY2025 data (released July 15, 2026) and changed the title from "Record 1,055" to "Pass 1,000 for the First Time (FY2024)."
Why we got it wrong Three days before this article was published (July 18, 2026), the Ministry of Health, Labour and Welfare had released the FY2025 "Status of Work-Injury Compensation for Death from Overwork and the Like" (July 15, 2026), in which mental-disorder awards reached 1,082, already surpassing the FY2024 figure the article treats. The cause was a failure to check the latest release of a published statistic and presenting the older year's data as a record.
Removed "15% were in favor" from the economist survey. The figure does not appear in the cited source.
Why we got it wrong On checking the source (the Japan Center for Economic Research economist survey), the breakdown of the 70% opposition is published, but the share in favour is not stated anywhere on the page. We removed the unsupported figure and replaced it with the breakdown we could verify.
Corrected the annual private high school learning cost and the share covered by the tuition support grant.
Why we got it wrong MEXT revised the figures in its "FY2023 Survey on Children's Learning Costs" on 16 January 2026 (the document is marked "replaced 16 January 2026 due to numerical correction"). The private high school annual total changed from ¥1,030,283 to ¥1,179,261. Because we used the superseded figure, the coverage share was overstated as well.
Reported by Found while auditing every article citing this statistic, prompted by a reader's report on a related article.
Updated the status of the amendment bill (from "aims for passage" to enacted on 31 March 2026, in force from April).
Why we got it wrong The article was published just after the cabinet decision and never reflected the bill's subsequent passage. We updated it while verifying facts for the statistical correction.
Corrected the three-year public-private learning cost gap from ¥1.29 million to ¥1.74 million.
Why we got it wrong MEXT revised the figures in its "FY2023 Survey on Children's Learning Costs" on 16 January 2026 (the document is marked "replaced 16 January 2026 due to numerical correction"). The private high school annual total changed from ¥1,030,283 to ¥1,179,261, and the public figure from ¥597,752 to ¥596,954. This article was published on 20 March 2026 — two months after that revision. We failed to re-check a published statistic and used the superseded figures. We also added an explanation that these survey figures are net of the tuition support subsidy, and standardised the three-year totals on the survey's own grade-by-grade figures (Table 2), which had previously been mixed with a simpler annual-times-three calculation.
Reported by Brought to our attention by a reader. Thank you.
Research integrity
