Your deck makes 40 claims. How many can you cite?

Open your deck. Read it one sentence at a time and mark every sentence that asserts something about the world: your growth rate, your market size, what customers say, what your product does, what happens next year. Most seed decks land somewhere between 30 and 50 marked sentences.

Now go back through the marked ones and ask a narrower question of each: if an investor stopped on this line and said show me, could you produce the exact document, the exact page, the exact cell? Not a folder. Not a spreadsheet that contains the number somewhere. The passage.

For most founders the second number is a fraction of the first. That gap is where diligence goes wrong, and it rarely goes wrong the way founders expect.

Key Takeaways

• A deck is a list of claims. Diligence is the process of testing them one at a time, and it moves at the speed of your slowest substantiation.
• The costly failure is usually not a lie. It is two true-ish numbers that disagree, or a real number nobody can locate.
• In January 2025 the SEC fined GrubMarket $8 million for giving investors financials it "should have known" were unreliable, while using a different, better set internally. The finding was inconsistency, not fabrication.
• Different claim types need different proof. Treating a projection like a measured fact costs you credibility on the facts too.
• Attaching a document is not substantiation. Substantiation is a claim bound to a specific passage that a human has confirmed supports it.

The expensive failure is inconsistency, not dishonesty

Founders imagine the catastrophic diligence outcome as getting caught in a lie. That does happen, and when it does it is spectacular. In June 2024 the SEC charged Ilit Raz, founder of the AI recruiting startup Joonko, with defrauding investors of at least $21 million. The complaint alleges that Raz misstated customer counts, candidate numbers, and revenue, supplied fabricated testimonials, and, when one investor grew suspicious, produced falsified bank statements and forged contracts (SEC press release 2024-70, June 11, 2024).

That is not the failure mode most founders need to worry about. Nobody reading this is forging bank statements.

The one worth worrying about is quieter. In January 2025 the SEC announced settled charges against GrubMarket, a private company, over its Series D. The order found that between November 2019 and February 2021, GrubMarket raised about $80 million while sending prospective investors financial information that overstated historical revenue by roughly $550 million over five years. At the same time, it was using a different set of figures, reflecting significantly lower revenues, for other corporate purposes including its tax filings. GrubMarket settled without admitting or denying the findings and paid an $8 million civil penalty (SEC press release 2025-18, January 17, 2025).

Read the SEC's framing closely. Mark Cave, an Associate Director in the Division of Enforcement, said the company "provided investors with financial information that painted a misleading picture of the company's historical performance, while at the same time using higher-quality financials for other business purposes."

The operative phrase in the order is should have known. That is what happens when a company holds two sets of true-ish numbers for two audiences and never puts them side by side. It is not a character flaw. It is a filing-system flaw, and it is common in companies that grew faster than their record-keeping did.

Your deck and your tax return are both making claims about the same reality. So are your board update, your data room, and the answer you gave an investor on a call in March. Nobody is checking whether they agree, because checking is tedious and no single person owns all five documents.

Not every claim needs the same proof

The reason founders under-substantiate is that they treat "claim" as one category, get overwhelmed by the volume, and substantiate nothing. But claims are not interchangeable. Sorting them is what makes the work finite.

A claim is a single assertion about the world that an investor could test on their own. A citation is that claim bound to an exact passage in a named document, confirmed by someone who has read both. The difference between those two words is most of this article.

A useful split, the one FoundOS uses internally, has six types. They do not cost the same to defend:

The failure that damages credibility most, relative to how innocent it is, is mislabeling across this table. A founder presents a projection in the same visual register as an internal fact (same font, same slide, no hedge), and an investor who spots it stops trusting the facts too. You spent real credibility to make a forecast look stronger.

The fix is not hedging everything. Hedged decks are unreadable and read as weak. The fix is that each claim carries the kind of support its type requires, and says which type it is when that is not obvious.

Early access

Get the diligence question bank when FoundOS opens.

We are building the workspace this post describes: 973 investor questions, claim-level evidence, and a readiness score that shows its own arithmetic. Join the waitlist for early access.

Email addressJoin the waitlist

A folder is not a citation

Most data-room advice misses this distinction entirely.

The standard guidance says: organize your data room, label your folders, make the cap table easy to find. That is real advice and you should follow it. But it operates at the level of documents, and diligence operates at the level of assertions.

An investor does not ask "do you have a financial model." They ask "why does your CAC drop by 40% in Q3." The answer to that lives in one row of one tab of one file, plus a reason. Handing over the whole model is not answering. It is delegating your own homework to someone evaluating you.

So the unit that matters is a claim bound to a specific passage: the document, the exact excerpt, the locator (page, section, cell), and a human confirmation that the passage actually supports the claim. That last part is not ceremony. A quote can be real, correctly transcribed, and still not support the sentence it has been attached to. Only a person who has read both can attest that it does.

Two things become visible fast:

Claims with nothing behind them. Usually numbers that were true once, cited from memory, and have quietly drifted. Nobody lied. The number just aged.

Claims where the evidence says something adjacent. You claimed customers churn over onboarding friction. The transcript says one customer, once, mentioned onboarding among four complaints. That is a real signal and a weaker claim than the one on the slide.

Both are findable in an afternoon. Both are much cheaper to find in your own afternoon than in someone else's diligence call.

Do this before your next raise

Four passes over the deck you already have. This is not a maturity model.

1. Count. Read every slide and mark every sentence asserting something about the world. Do not fix anything yet. Founders underestimate this number, and the count alone changes how the deck reads.

2. Type each one. Internal fact, external fact, customer evidence, projection, hypothesis, opinion. Two minutes a claim. Any claim you cannot type with confidence is a claim you have not thought through, and that ambiguity is itself the finding.

3. Locate the proof. For each fact-type claim, write down the document and the passage inside it. Timebox to five minutes each. Anything you cannot locate in five minutes, an investor will not locate either. Mark it.

4. Cross-check the overlaps. Take every number that appears in more than one place (deck, board update, model, data room, tax filing) and put the versions side by side. This is the GrubMarket pass, and it is the one everybody skips because it is boring and requires no judgment. It is also the one with an $8 million case attached to it.

What comes out is not a fixed deck. It is a list, sorted into claims that hold, claims needing better evidence, and claims to cut. Cutting is a legitimate outcome and usually the fastest one. A deck with 25 substantiated claims beats a deck with 40 claims where six collapse under a follow-up question.

The order of operations most founders get wrong

The instinct is to build the data room first and worry about substantiation when someone asks. That inverts the actual dependency. The data room is storage. Substantiation is the work. A beautifully organized room full of documents that nobody has mapped to specific assertions is a filing cabinet with good labels.

Do it the other way. Start from the claims, because the claims are what get tested. Let the list of claims tell you which documents you need. You will usually find you need fewer than you thought, and that three of them do not exist yet.

That inversion is the whole idea behind what we are building. It holds a bank of 973 investor questions across 109 modules, and answers them in three layers: the investor-facing conclusion, the analysis behind it, and the diligence notes. Evidence attaches at the level of the individual claim rather than the folder. Readiness is reported as four things you can inspect (coverage, evidence, consistency, review) rather than one number you have to trust. It measures how prepared you are. It does not predict whether you will raise, and anything claiming to do that is selling you something.

But the exercise above needs none of that. It needs your deck, a couple of hours, and a willingness to write "cannot locate" next to a number you have been saying out loud for six months.

Go count. The number will be higher than you think, and the gap will be wider than you want. Better today than on the call.

Frequently asked questions

How many claims does a typical seed deck make?

Most seed decks contain 30 to 50 sentences that assert something testable about the world. That range comes from counting marked sentences across decks, not from a published study, so count your own rather than trusting the range. The count matters less than the ratio of claims you can locate proof for.

What counts as substantiation in investor diligence?

Substantiation is a specific claim bound to an exact passage in a named document (page, section, or cell) with a human confirmation that the passage supports the claim. Attaching a whole document is not substantiation, because it leaves the reader to find the supporting passage and decide for themselves whether it says what you said it says.

Is it dishonest to put projections in a fundraising deck?

No. Projections belong in a deck. The problem is presenting one in the same visual register as a measured fact with no signal of which it is. Investors expect forecasts and discount them accordingly; what they penalize is having to work out for themselves which numbers were measured and which were modeled.

What is the most common diligence gap at Series A?

Practitioner accounts point to missing IP assignment agreements from founders and early contractors. One diligence write-up calls it "the single most common gap that experienced investors identify," and notes that investors typically open a data room with the cap table and financial statements first, because "these two documents tell investors whether the deal's foundations are solid before they spend time on anything else" (Pitchwise, retrieved 2026-08-06). Treat that as practitioner observation rather than measured data. It is a consultancy's account, not a study. These are document-level gaps rather than claim-level ones, and worth closing separately from the exercise in this post.

Can I do this without a tool?

Yes. The four passes above need a copy of your deck, a spreadsheet, and roughly two hours. Tooling helps when the claim count grows past what one person can hold in their head, and when the same claims have to stay consistent across a deck, a model, a board update, and a data room over months.

---

How this post was sourced. It would be a poor look to argue for citation discipline and then not practise it. Both enforcement cases are cited to SEC press releases read in full at sec.gov, not to secondary coverage of them; the figures, dates and quotes above were checked against those pages. The practitioner claim about IP assignments is quoted from the page it links to and labelled as observation rather than data. The claim-type model and the question bank figures are our own, first-party, and stated as such. No statistic in this post is included without a source you can open. Corrections: [email protected].