BRENNEN

Industries / Finance and business services / Finance and banking / Private equity and venture capital

Valuation

Valuation is the work of putting a price on shares and loans in private companies that trade on no market, when an investor buys in and again every quarter after.

7

Milestones achieved, from 2006 to 2025.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2006

    One meaning of fair value

    US fund law has required since 1940 that holdings with no market price be valued at "fair value as determined in good faith by the board of directors". In 2006 the Financial Accounting Standards Board (FASB), which sets US accounting rules, issued SFAS No. 157. It gave fair value a single definition and sorted every value into three levels; Level 3 is for values that rest on "significant unobservable inputs", the holder's own assumptions, which is where holdings with no observed price fall. It took effect at the start of 2008 for all reporting companies, by the SEC staff's account.

    law.cornell.edusec.gov

    Credit

    the Financial Accounting Standards Board

  2. 2007

    A price for startup options

    Under section 409A of the US tax code, a stock option escapes the tax rules for deferred pay only if its exercise price is never below the share's fair market value on the day it is granted. The Treasury's final regulations of 17 April 2007 presume a private company's value is right, unless it is "grossly unreasonable", when it comes from an independent appraisal no more than 12 months old or, for "illiquid stock of a start-up corporation", from a written report by someone with, in general, at least five years of relevant experience. The start-up rule does not apply if a sale of the company is expected within 90 days or a stock market listing within 180 days. These rules are why US startups buy a "409A valuation" before granting options.

    govinfo.gov

    Credit

    the US Treasury and the Internal Revenue Service

  3. 2013

    Invest now, price later

    The SAFE (simple agreement for future equity) was "created at Y Combinator by Carolynn Levy in 2013". It lets an investor "fund your startup now in exchange for the right to shares of stock in your startup later", so the investor's shares are issued at a later financing. Y Combinator "standardized on the post-money SAFE in 2018" and calls the SAFE "the most common way early-stage startups raise their early investments".

    ycombinator.com

    Credit

    Carolynn Levy and Y Combinator

  4. 2017

    Headline prices overstated

    A startup's headline "post-money valuation" multiplies all its shares by the price of the newest preferred shares, though those shares often carry protections the others lack. Will Gornall and Ilya Strebulaev valued 135 US unicorns, private companies with reported valuations above $1 billion (a term Aileen Lee of Cowboy Ventures used in 2013, when she counted 39 US software companies), one share class at a time, from terms in their legal filings. In their NBER working paper of October 2017, reported valuations averaged 50% above fair value, common shares were 58% overvalued, and 65 of the 135 fell below $1 billion. Square's Series E price implied $6 billion; their model gave $2.2 billion.

    LessonA round's price per share values only that round's shares, not the whole company.

    nber.orgtechcrunch.com

    Credit

    Will Gornall (University of British Columbia) and Ilya Strebulaev (Stanford)

  5. 2019

    A $47 billion price unwinds

    WeWork's valuation "was set at $47bn in a January fundraising from SoftBank". When banks tested demand for a listing, public market investors "balked at a valuation as low as $15bn", and WeWork withdrew its IPO plan on 30 September 2019. On 23 October SoftBank agreed a rescue, buying shares at $11.60 and offering $19.19 to other holders, that would leave it with about 80% of the company; press reports said the deal valued WeWork below $8 billion.

    LessonA price set by one large private investor is not the price public buyers will pay.

    irishtimes.comaljazeera.comgroup.softbank

    Credit

    WeWork and SoftBank Group, the parties; The Irish Times and Al Jazeera reported the valuations

  6. 2022

    Klarna's price falls 85%

    Klarna, a Swedish buy-now-pay-later lender, raised $800 million in July 2022 at a $6.7 billion post-money valuation, against $45.6 billion in June 2021. This was a down round, a funding round at a lower price than the one before. Klarna said its listed "peers are down 80-90% vs peak valuations" and that its cut was "on par with its public peers". Its prospectus records $799 million received from the July 2022 round, and it listed on the New York Stock Exchange in September 2025 at $40 a share.

    LessonA private company's price can stay above its listed peers' until it next raises money.

    theblock.cosec.gov

    Credit

    Klarna; reported by The Block

  7. 2025

    Guidelines for private capital

    The International Private Equity and Venture Capital Valuation (IPEV) Board, whose founding associations were AFIC, BVCA and Invest Europe, published a new edition of its valuation guidelines on 11 December 2025, in effect for quarterly reporting periods beginning on or after 1 April 2026; 40 associations endorse them. They cover venture, buyout, infrastructure and credit holdings and are written to meet both IFRS and US accounting rules. They require calibration: checking that a valuation model, fed the market inputs of the day the investment was made, reproduces the price actually paid, then carrying those inputs forward. They also state that "significant valuation differences for identical securities among investors in the same underlying portfolio company may indicate weakness in valuation judgement or process".

    privateequityvaluation.comprivateequityvaluation.comprivateequityvaluation.com

    Credit

    the IPEV Board

Watch for

  1. Exit prices above the marks

    A study published by Invest Europe, the European industry association, found that listed private equity firms sold direct investments at prices averaging 36% above their last balance-sheet values, and funds of funds at 47% above, across 193 disposals of direct investments and more than 3,000 fund-of-funds exits from 2019 to 2024. Its authors read this as a sign that valuations "tend to be conservative rather than aggressive".

    Achieved whenA party outside the industry repeats the test from the firms' reports, counting holdings written off or still unsold as well as those sold, and finds a similar gap.

    dbag.com

    Claimed by

    Invest Europe, December 2025, as announced by Deutsche Beteiligungs AG, a listed private equity firm

  2. Forecasting loan marks between filings

    Syndicate, an open-source project on GitHub under the MIT license, matches 15,951 private loans held by 189 funds, using the loan-by-loan values that business development companies (BDCs, US funds that lend to private companies) file with the SEC, from September 2022 to June 2026, and estimates each loan's value on days between filings. In its own out-of-sample test, where another lender had already filed a loan, its model removed 56% of the squared error of simply repeating the last value, while a simple rule an analyst could apply by hand removed 58%.

    Achieved whenAn outside party scores its forecasts for the third quarter of 2026, which its README names as the first clean test, and publishes the result.

    github.com

    Claimed by

    the Syndicate project, October 2026

  3. Fund marks side by side

    US mutual funds that own shares in private companies must report a value for each holding on Form N-PORT. BriMindInvest, a stock-analysis website, publishes these values side by side, each linked to its filing: its snapshot of 14 November 2026 holds 9,657 marks of 47 companies by 148 funds for the second and third quarters of 2026, with the spread between funds for each company and the whole dataset free to download. It shows only companies held by at least three funds from two unrelated managers, reads a sample of filings, and calls its counts "a floor". An earlier study, made by hand from filings for 2010 to 2016 before N-PORT existed, found prices for the same security differed by 10.0% on average across fund families.

    Achieved whenAn outside party checks a random sample of its marks against the filings, and the record runs back through earlier quarters, including the markdowns of 2022.

    brimindinvest.comabfer.org

    Claimed by

    BriMindInvest, November 2026

Open

  1. 01

    Preferred terms from public charters

    An open dataset of the rights attached to each class of a startup's preferred shares, read from the charter (certificate of incorporation) that US companies file with the SEC as an exhibit to their listing prospectus, which shows the company's terms just before it went public. For each series, record the original price, the liquidation preference (how many times its money a class gets back first in a sale), whether it also shares in what is left, its rank against other classes, its anti-dilution formula (how its conversion price adjusts if later shares sell cheaper) and any guaranteed return at a listing. Each field links to the exhibit and section it came from. It records filed terms and values no company.

    The brief
    Credit · unclaimed
  2. 02

    Share-class math, scored

    A public test set that scores AI models and software on startup share math. Each case is an invented company whose cap table, the list of who owns which shares, is written in Open Cap Format. The companies hold common shares, options, SAFEs, convertible notes and several classes of preferred shares with different rights, such as seniority, participation and IPO guarantees. Questions ask who owns what after the next priced round, and how much each class receives if the company is sold at several prices. Answer keys come from two independent open calculators and a hand check. Every company is invented, so no real company is valued.

    The brief
    Credit · unclaimed

Open 01 · the brief

Preferred terms from public charters

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open dataset of the rights attached to each class of a startup's preferred shares, read from the charter (certificate of incorporation) that US companies file with the SEC as an exhibit to their listing prospectus, which shows the company's terms just before it went public. For each series, record the original price, the liquidation preference (how many times its money a class gets back first in a sale), whether it also shares in what is left, its rank against other classes, its anti-dilution formula (how its conversion price adjusts if later shares sell cheaper) and any guaranteed return at a listing. Each field links to the exhibit and section it came from. It records filed terms and values no company.

Why it is still open

Gornall and Strebulaev read such terms for 135 unicorns for their 2017 study. The largest study found, Spencer Williams's 5,564 venture financings from 2004 to 2015, used records that one law firm, unnamed, kept for its clients and anonymized before handing over. Law firms publish only the share of deals with each term: the Canadian firm Osler, for example, found a senior liquidation preference in 20.4% of the Canadian rounds in its 2025 data, and cites Wilson Sonsini's 21% for US financings. The charters themselves are public: HashiCorp's, filed with its S-1 in November 2021, sets out its preferred shares' liquidation preference and conversion price adjustments.

fordhamcorporatejournal.orgosler.comsec.gov

Built with what exists

EDGAR full-text search to find charter exhibits filed with S-1s; the National Venture Capital Association's model certificate of incorporation as the field template; an agent that copies terms into fixed fields, such as Meta's Muse Ember (see the Note "The jobs agents repeat"); Python.

  1. First step

    Pull the charter exhibits from the S-1s of ten companies that listed in 2021 and fill the fields by hand.

  2. Done looks like

    Terms for every preferred series of at least 300 venture-backed US companies that listed from 2010 to 2025, each field linked to its section, with code and data public; a second person checking 50 random charters and agreeing on at least 95% of fields; and published shares of companies with senior preferences, participation or a guaranteed listing return, set beside Gornall and Strebulaev's figures.

Open 02 · the brief

Share-class math, scored

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test set that scores AI models and software on startup share math. Each case is an invented company whose cap table, the list of who owns which shares, is written in Open Cap Format. The companies hold common shares, options, SAFEs, convertible notes and several classes of preferred shares with different rights, such as seniority, participation and IPO guarantees. Questions ask who owns what after the next priced round, and how much each class receives if the company is sold at several prices. Answer keys come from two independent open calculators and a hand check. Every company is invented, so no real company is valued.

Why it is still open

Gornall and Strebulaev showed that such terms move the value of each share class by large amounts, yet the open tools found are calculators, not graded tests. The nearest benchmarks score other tasks: a December 2025 paper treats checking a cap table against signed legal documents as a benchmark for AI agents and links no data, and WaterfallBench, released in November 2026, scores models on the payment order of car-loan bonds, not startup shares.

arxiv.orghuggingface.co

Built with what exists

Open Cap Format, the Open Cap Table Coalition's data standard; open calculators such as venture-dealr and cap-table-tool; Python; models to test such as Claude Fable 5.2 (see the Note "Three working devices") and the open-weight Mistral Large 4 (see the Note "A trillion open parameters").

  1. First step

    Write 20 invented cases in Open Cap Format, run them through two open calculators, and list every case where the calculators disagree.

  2. Done looks like

    At least 300 cases from simple to hard, answer keys on which both calculators agree to the cent, published scores for at least five models, and a second person reworking 30 random answers by hand with no differences.