Industries / Construction and property / Real estate
Appraisal
Appraisal is the work of estimating what something is worth, whether a home, an office building, a company or a share in one, a machine, a vehicle or a work of art, and of writing the estimate up so that a lender, buyer, court, tax office or estate can rely on it and check it.
1932
The National Association of Real Estate Boards, the US realtors' federation, turned its Appraisal Division into a separate body, the American Institute of Real Estate Appraisers, which adopted standards of practice and a code of ethics and certified experienced appraisers with the MAI designation.
38
Milestones achieved across 7 domains.
21
To watch for: announced, not yet confirmed outside.
17
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Land and goods were valued for taxes and loans long before valuation was a profession. It became a trade of its own in the early 1900s: London surveyors worked from printed tables that turned a yearly rent into a capital value, US appraisers formed their own institute in 1932, and a society founded in 1936 took in appraisers of machinery and other movable goods. The written rulebooks came after crashes. Britain's Red Book of valuation guidance followed the property collapse of 1972–73, and US appraisal standards and state licenses followed the savings and loan crisis of the 1980s.
Where it stands
The US national registry held 89,618 appraiser credentials at the end of 2025, which the federal Appraisal Subcommittee estimates belong to about 65,857 people, down from 121,407 credentials in 2008. A full appraisal is still the norm for US home purchases: it backed 77.6% of Freddie Mac's and 85.7% of Fannie Mae's purchase loans in February 2026, by the American Enterprise Institute's count, and since 2 November 2026 new reports for those loans must use one redesigned form. Values still lag and still get disputed: US downtown office appraisals fell 43% from their peak by mid-2024 while sale prices fell 51% (MSCI's figures), researchers disagree on how far race moves home appraisals, and in fiscal 2023 the IRS's art panel cut artworks claimed at $150.5 million to $81.4 million.
What is open
Firms that sell automated and AI valuations mostly publish only their own accuracy figures, and most of the open jobs below are the independent tests and public datasets that would check them. Several are sealed tests that score estimates against prices nobody knew when the estimates were locked: takeover prices, auction hammer prices and home sales not yet recorded. Others turn public records into data, such as leases and loan valuations filed with the SEC, German land values, the IRS art panel's reports and state boards' disciplinary orders, or build made-up appraisal reports with planted faults to score review software. AI models that read long documents, and agents that collect pages from public websites, put this work within reach of a student or a small team.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Homes
Home appraisal is estimating what a house or flat would sell for, most often so that a lender knows the value of the home behind a mortgage, by looking at the home and comparing it with recent sales of similar homes nearby, called comparables.
4 · 3 · 2
Last achievedOne report replaces the forms2026
Watch forCommittees require sale details
2
openCommercial property
Commercial property valuation is estimating what offices, shops, warehouses, hotels, blocks of rented flats, data centers and land would sell for, mostly by turning the rent a building earns into a capital value, for lenders, investors, property funds and company accounts.
6 · 3 · 2
Last achievedOffice appraisals trail sale prices2024
Watch forAI drafts appraisal reports
2
openBusinesses
Business valuation is estimating what a company, a block of its shares or an intangible asset such as a brand or patent is worth, for taxes, estates, divorces, lawsuits, deals and company accounts.
5 · 3 · 3
Last achievedDeal price beats the model2017
Watch forClaude builds cash flow models
3
openEquipment and art
Equipment and art appraisal is estimating what movable things would sell for, such as machines, vehicles, aircraft, artworks, jewelry and collectibles, mostly from price guides, auction results and dealers' sales, for lenders, insurers, tax offices, courts and estates.
6 · 3 · 3
Last achievedAuction house appraisal fails2025
Watch forAI values art for lenders
3
openAutomated valuation
Automated valuation is computer models that estimate what a property is worth from records of its features and of nearby sales, with nobody visiting, for lenders, tax offices and property websites.
5 · 3 · 2
Last achievedTesters catch springing estimates2024
Watch forIAAO's new fairness test
2
openBias and review
Bias and review is the checking of appraisals after they are written, by a second appraiser, by lenders' and loan buyers' software, and by owners who ask for a reconsideration of value, and the evidence and law on whether values are pulled by race or by other forces, such as the agreed sale price.
6 · 3 · 2
Last achievedFederal push reversed2025
Watch forContesting a value becomes law
2
openStandards
Standards are the rulebooks and gatekeepers of valuation: the documents that say how a value must be worked out and written up, and the institutes and licensing boards that decide who may sign one.
6 · 3 · 3
Last achievedUSPAP guidance covers AI2026
Watch forIVS rules for AI tools
3
open
Advances to build with
4 research notes

Claude Fable 5.2
Three working devices
Anthropic says its AI model Claude Fable 5.2 designed three prototype devices that worked without electrical modifications, with no outside test reported.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.

Cursor Agent SDK
Findings that cite notes
Cursor released a kit for building AI agents, and its example, a code reviewer that cites the note behind each finding, is known only from Cursor's account.

Microsoft ThinkingBox
All twenty times
By Microsoft's own count, 50 steps of training took an AI model from passing 38 of 507 test tasks on all 20 tries to passing 107.
Open in appraisal
3 of 17 open deliverables
Open · Homes
German land values, one file
A single file of Germany's current standard land values for residential zones, merging each state's official open data into one format, with the license and required attribution recorded for every row and a list of the states whose terms do not allow reuse.
Built with what exists
State open-data downloads, such as Brandenburg's land-value files under the Datenlizenz Deutschland attribution license (dl-de/by-2-0); Python with GeoPandas; QGIS for spot checks; Browser Use 1.0, an open-source browser agent, to find each state's download page (see the Note "Deleting 80% to start over").
Open · Commercial property
Lease terms a valuer needs
An open set of 200 commercial leases (office, retail and industrial) taken from exhibits to US SEC filings, each labeled for the terms a valuer types into a cash flow model: parties, floor area, start and expiry dates, rent for each year, the rule for rent increases or reviews, rent-free months, break and renewal options, who pays the running costs (a gross or a net lease) and the deposit. Ship it with a scoring script.
Built with what exists
EDGAR full-text search, which covers all electronic SEC filings since 2001, exhibits included; Meta's Muse Ember for long-document extraction (see the Note "The jobs agents repeat") and Claude Fable 5.2 (see the Note "Three working devices") as baseline readers; Gemini Decision 1 for multiple-choice fields such as the lease type (see the Note "A thousand questions, one pass"); release under CC BY 4.0, as CUAD was.
Open · Businesses
Sealed pre-deal valuation test
A public test in which AI agents value companies that were later bought for cash. Each agent sees only filings and share prices from before the deal was announced, and locks a value per share and a range, which are hashed and time-stamped. The deal price is revealed only afterward.
Built with what exists
SEC EDGAR full-text search and its financial-data interfaces; Aswath Damodaran's free cost-of-capital and industry multiples data at NYU Stern, last fully updated in January 2026; agents built on models such as Claude Fable 5.2 (see the Note "Three working devices"), each case run many times (see the Note "All twenty times"); deals announced after each model's training cutoff, or with names and dates masked, so no model can recall the answer.
Where it sits
Broader
In outline