August 2026: Verified figures, statutory acquisition costs, and analyses fully in your language
Every analysis now passes through a validation layer before it reaches you: currency conversions are re-derived from the authoritative rate, yields are reconciled across the narrative, the tables and the PDF, and acquisition costs are computed from Japan's statutory schedules for your property's actual price. Analyses render completely in your chosen language, a new section sets out what the analysis could not determine and which licensed professional to ask, and condominiums in Japan's major developer brand series are recognized by name.
April was about data, May about personalization, June about judgment, July about risk and location. August was about precision. As the analysis has taken on more of the work an investor would otherwise do by hand, the standard it has to meet has risen with it: every figure needs to be verifiable, reproducible, and consistent wherever it appears. So this month went into the machinery behind the report, a verification layer that sits between the calculations and the page, statutory computation for the costs that depend on Japanese law, and a single authoritative record behind every analysis. Here's what shipped.
1. A verification layer between the calculations and your report
Tatemono IQ computes the figures that matter and hands them to the AI to write about. August added a layer that verifies the finished report against those calculations before it is saved.
Currency conversions are re-derived. Every ¥X (approximately $Y) pair in an analysis is parsed back out of the finished text and recomputed from the same authoritative exchange rate the analysis was generated with. Anything outside a rounding tolerance is corrected before the analysis is persisted, so your report, your PDF export and any link you share all carry the same verified figures.
Yields are reconciled everywhere they appear. Reconciliation now covers the narrative prose, the summary tables and the PDF together, so the yield quoted in a sentence and the yield in the table are the same number, and the narrative is checked against the analysis's own calculations block.
Unknowns are reported as unknown. Where a figure genuinely cannot be derived from the property record, the analysis says so instead of presenting a placeholder as a result. An expense ratio is reported only when gross rent is known, and an unknown net yield is carried through as unknown rather than being scored.
Price-anomaly checks run on every path. The checks that catch data-entry errors in price now run on every route that generates or regenerates an analysis, and are correctly scoped for sectional-ownership units (区分), where a building's total unit count (総戸数) describes the building rather than the holding being analyzed.
2. Acquisition costs, computed from Japan's statutory tables
Acquisition costs are now computed on the server from the statutory schedules, for your property's actual price: the stamp duty (印紙税) bracket that applies at that price, and the brokerage commission under 宅建業法, including the ¥330,000 tax-inclusive ceiling for properties under ¥8M introduced by the 2024 reform.
The panel you see is populated from that same statutory computation, so the figures in front of you are the correct ones for the property before you touch anything. If you do want to model different assumptions, the system now tracks precisely which fields you edited: your overrides apply, and every field you leave alone stays on the statutory basis.
Source: Japan statutory acquisition-cost schedules (印紙税 price brackets, 宅建業法 brokerage commission rules incl. the July 2024 low-price reform)
3. One authoritative property record behind every analysis
An analysis can be started several ways: from the property page, from a list, in a batch. All of them now assemble their data from a single authoritative source, so every analysis reads the complete property record (building structure, storeys, unit count, floor area, zoning and the rest) regardless of how it was triggered. The same property gives you the same analysis, and a build-time guard now proves that every field the report can display has a producer behind it, so the report and the data stay in step as both grow.
Converging on one record also brought the demolition-cost estimate into service: it now renders for the properties it was designed for, land-only listings and detached houses past the age threshold.
The Property Scorecard gained the same reproducibility. Its Financial Returns rating is now scored against explicit numeric yield bands rather than a qualitative judgment, so identical figures produce an identical star rating every time you generate an analysis, in every language.
4. A new section: What This Analysis Could Not Determine
Every analysis now closes with a section titled What This Analysis Could Not Determine, covering three things:
- Data that was not available, named as specific fields rather than a general note about limited data.
- Figures that depend on an assumption you did not supply, with the assumption restated next to the figure.
- Questions for a licensed professional, each matched to the right profession: a certified real estate appraiser (不動産鑑定士) for valuation, a certified tax accountant (税理士) for tax treatment, a judicial scrivener (司法書士) for title and registration, a licensed real estate transaction agent (宅地建物取引士) for transaction disclosures, and an administrative scrivener (行政書士) for permits and licensing.
This section replaces the analysis's closing recommendation, and the change is deliberate. Tatemono IQ is a research and analysis tool, not a licensed investment adviser and not a licensed appraiser (不動産鑑定士). The figures, the risk profile, the market context and the side-by-side comparison of strategies are all still there, and they are sharper than they have ever been. What sits at the end of the report now is the set of open questions to take to the right professional, which is a more useful place to finish than a verdict. The guided tour was updated to match.
5. Major developer brand series, recognized by name
For condominiums, the developer behind the building is a genuine quality signal, and in Japan it is usually legible from the building's own name. The analysis now matches a property against a curated list of the major developers' branded series: Mitsui Fudosan's Park Homes (パークホームズ), Park Court and Park Tower; Mitsubishi Estate's The Parkhouse (ザ・パークハウス); Sumitomo Realty's City House and City Tower (シティタワー); Nomura Real Estate's Proud (プラウド); Tokyu Fudosan's Branz (ブランズ); Daiwa House's Premist (プレミスト); and Sekisui House's Grande Maison (グランドメゾン).
When a match is found, whether in the property name, the developer field, the management company or the constructor, the analysis and the synopsis name the series and treat it as a positive factor for construction quality and professional management. Two deliberate constraints keep it trustworthy: matching is exact against a curated list rather than a fuzzy guess, so an unrelated company that happens to share a word is never mistaken for a brand series; and where there is no match, the analysis is explicitly instructed not to speculate about the developer's or management company's reputation. Silence is the honest answer there, not an invented one.
Source: Curated major-developer condominium brand series, researched and maintained in-house
6. Analyses that render completely in your language
Localization now reaches the structural furniture of the report, not just the prose. The Financial Summary table's column headers and row labels, the entire Property Scorecard table, and the Property Assessment block all render in the language you asked for, in the app and in the PDF export, across Japanese, Korean, Simplified Chinese and Traditional Chinese. Traditional Chinese output is additionally checked for script purity on save, so Simplified characters cannot slip into a zh-TW analysis.
Follow-up questions were rebuilt on structured output. Rather than being parsed out of free-form text, the questions and their supporting metadata now come back as structured data, which makes extraction reliable in every language and makes the feature straightforward to extend. Questions are scoped to one topic each where that reads more naturally to a Japanese counterparty, and existing non-English analyses were reprocessed so they carry their follow-up questions too.
7. Start on your phone, continue at your desk
The core workflow, importing a listing with the browser extension or uploading a Maisoku (マイソク) PDF, is built for the desktop. Signing up on a phone now bridges to that instead of stopping at it.
New accounts opened on a phone see a one-time screen explaining what the desktop experience covers and what you can explore right now, with a single tap to email yourself a link to continue on desktop. You can also dismiss it and carry on browsing the demo analyses on your phone. It appears once per account, in all five languages, and never on desktop.
Also improved in August
- More control over analyses in flight. A Cancel option is available for any analysis that has not yet finished, not only long-running batch jobs. Quota is returned automatically when a job does not complete, and a submission that your subscription does not currently permit returns a clear explanation of why.
- Aki reads your whole result set. Ask about your search results at any page size and Aki answers from all of them, with counts, cities, and price and yield ranges drawn from your actual visible set. Exchange-rate figures stay accurate throughout a long conversation and in every display currency.
- Fresher, cleaner underlying data. Production station ridership is now on the current MLIT survey year, with figures for 262 more stations. The open-data amenity layer is filtered to Japan at the point of extraction, removing roughly 195,000 points that fell inside the extraction bounds but outside the country. POI refreshes now verify index parity as part of the run.
- Maisoku export reflects your own relationship to the listing. The transaction mode (取引態様) printed on an exported flyer describes the exporting team's own position, and the export flow notes that redistributing another firm's listing requires that firm's permission. On the property form, transaction mode is validated against property status before you save, and the analysis is told when a record is internally inconsistent.
- Presentation polish across the report: approximation symbols render as intended, prefectures appear as their names alone, and the disclaimer on list analyses is scoped to the data those analyses actually use.
Source: MLIT S12 station passenger survey; Overture Maps + OpenStreetMap + MLIT amenity data