# Category bet vs company endorsement: a guide | SourceX

Canonical: https://refer.sourcex.si/guides/the-difference-between-a-category-bet-and-a-company-endorsement
Last updated: 2026-11-05

A category bet is the fund's own claim about where demand is heading. A company endorsement is a claim that a particular company, or SourceX itself, is a fit, a good choice or safe. The two need different evidence and different owners. Before any thesis sentence reaches founders or LPs, tag its level (category, company, program or outcome) and its owner (the fund, a cited source, the CEO, SourceX's published terms, or nobody yet). A sentence that claims more than its owner can support is an endorsement in disguise: rewrite it or cut it.

Most slips are small. SourceX's screening guidance names the commonest one: a newsletter line that reads "our companies qualify" implies an acceptance nobody has given. The guide below stops that drift without silencing the thesis.

## Four levels a thesis sentence can claim

Every sentence a platform team writes about AI data sits at one of four levels. The level decides who may own the sentence and what evidence it needs.

| Level | What the sentence claims | Who can own it | Evidence that supports it | The usual slip |
|---|---|---|---|---|
| Category | Where AI developers' needs are moving, and which kinds of records fit them | The fund, as its stated view, or a cited publisher | Dated primary research, named in the sentence | A bet written as a fact |
| Company | What one company keeps, for how long, and whether it is the company's own | That company's CEO | The CEO's confirmation, dated, at metadata level | Inferred from sector, funding or a board pack |
| Program | What SourceX requires and pays | SourceX's published terms and screening criteria | The terms page and its version | Paraphrased into a promise |
| Outcome | Whether a company will be licensed and what anyone earns | Nobody yet | Only SourceX's review and a signed agreement decide it | Any forecast, however hedged |

The levels do not convert into one another: repetition never turns a category claim into a company claim, and a strong fit never turns a program fact into an outcome.

A sound category bet looks like this. Anthropic's engineering post on agent evaluations describes a task as a test with defined inputs and success criteria, and recommends grading what actually happened in the environment rather than what the agent reports (our paraphrase of the post). A fund could reasonably build a thesis on that: records showing how real cases ended are the kind evaluation work needs. That is a claim about kinds of records, silent on which company keeps them, whether they are its own and whether its CEO would discuss licensing them.

## From the bet to a question one CEO can answer

A category bet earns its keep when it tells you which companies to ask and what to ask them. Translate it in three moves.

1. **Name the record, not the sector.** "Cases tracked from intake to a recorded outcome, in one system, for several years" can be checked. "Logistics data" cannot.
2. **Name who would keep it for itself.** The company doing the work for its own business, not the vendor hosting the software or the outsourcer working on clients' files.
3. **Write the question a CEO can answer without opening a file.** "Does your company keep its own records of each case from start to final outcome, and since what year?"

That question is an ask, not a verdict, so nobody can read it as the fund vouching for the company or for SourceX. Choose which companies to ask by who owns the CEO relationship, not by who is most admired in the portfolio; venture backing is not among SourceX's checks. See also the [venture platform overview](/for/venture-platform-teams) and the [thesis-to-introduction canvas](/guides/from-ai-data-thesis-to-a-useful-company-introduction).

## The thesis attribution language guide

The guide pairs sentences platform teams actually write with the level each claims, the owner it needs and an attributed version that keeps the meaning without the endorsement. Use it on newsletters, LP letters, founder emails and slide notes.

| What a platform team wants to say | Level it claims | Owner it needs | Attributed version |
|---|---|---|---|
| "Business records are the next big AI asset." | Category, stated as fact | The fund, as a view | "Our working thesis, dated this quarter: AI developers will value records that show how real work ended. Here is what we read and what would change our mind." |
| "Our portfolio is sitting on valuable data." | Company, for every company at once | Each CEO | "A few companies we know may keep years of their own operating records. We have not checked, and only each CEO can say." |
| "Logistics companies qualify for SourceX." | Program and company, merged | SourceX's criteria, then the CEO | "SourceX's published screen starts with 50+ full-time employees at peak, documented operating history, rights to license and an authorized contact. Sector decides nothing on its own." |
| "This company has fifteen years of claims data." | Company | That company's CEO | "Its CEO told us, on a date we record, that the company keeps claims records back to its founding year in one system." |
| "SourceX will pay you for your data." | Outcome | Nobody yet | "SourceX decides in its own review whether to license anything, and pays only under a signed agreement." |
| "We recommend SourceX." | Endorsement of a counterparty | The fund, with its connection disclosed | "We are a registered SourceX referral partner and may receive a reward if a company we introduce licenses data. We mention it as one option to look at." |
| "SourceX handles all the compliance." | Counterparty quality | Not the fund | "SourceX describes its rights review and contract on its own site; your CEO and counsel judge them." |
| "We're backing the AI data category." | Investment view | The fund's investment team | Keep investment views out of referral messages and send investment questions to SourceX's investor overview. |

Three rules sit under the table. Every sentence has one owner, named or obvious. Every sentence carries evidence at its own level. And a referral relationship travels with any favorable mention: where a platform publishes something that recommends the program, FTC staff guidance on the Endorsement Guides says a connection the audience would not expect, payment included, should be disclosed clearly and conspicuously (our paraphrase). Disclosure does not repair an overstated claim, so fix the level first. For wording at the moment of introduction, see [introducing without endorsing](/guides/introduction-without-endorsement) and [FTC disclosure examples](/resources/ftc-disclosure-examples-referral); associations face the same line in a [vendor endorsement policy](/questions/association-endorsement-policy-vendor).

## What must be confirmed before a company's name appears

A name in a thesis document reads as a selection. Collect four facts first, each with a date and the person who gave it.

| Evidence | Who gives it | What it lets you write |
|---|---|---|
| A relationship owner who will ask | The partner, LP or founder who knows the CEO | That you can raise the question, nothing more |
| The records exist, span years and sit in a named system | The CEO, or the records owner the CEO points to | A metadata description, attributed and dated |
| The records are the company's own | The CEO, subject to contracts only a rights review reads | That the company believes it holds them, never that rights are clear |
| The CEO wants the conversation | The CEO, who authorizes a licensing discussion and signs for the company | That an introduction may follow |

Whatever level a claim sits at, every row answers to the same published fit facts: a full-time headcount that reached 50+ at peak, contractors not counted, a documented operating history, rights to license, and an authorized contact able to approve the conversation. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Questions stay descriptive (what, how many, how far back, which system); samples, exports and logins never enter it. Whether the records match a current program, and whether customer contracts limit reuse, stays unknown until SourceX reviews the company.

## A platform paragraph, before and after (Illustrative)

Illustrative. A fictional newsletter paragraph and its rewrite; no real fund or company.

**Before.** "We believe the next frontier of AI is real operational data, and our portfolio is perfectly positioned. Companies like our 300-person freight brokerage are exactly what AI labs want, and through our partnership with SourceX, founders can turn their data into a new revenue line."

Tagged, it makes five claims: a category bet stated as belief (acceptable), a claim about every portfolio company (no owner), a claim about one named business (no CEO confirmation), a claim about what AI labs want (no source) and a revenue outcome (nobody can own it). "Our partnership" hides the reward.

**After.** "Our current thesis is that AI developers will need records of real work with recorded outcomes; the linked note explains why, and what would change our view. If your company keeps years of its own case records, SourceX runs a screen for direct data licensing, and its fit facts are public. We are a registered SourceX referral partner and may be rewarded if a company we introduce licenses data, so treat this as one option among others."

The rewrite keeps the thesis, drops the named company and the revenue line, attributes the program facts and discloses the connection.

## An eight-CEO check that can break the bet

A small test keeps the thesis honest in fact. Choose eight companies whose CEOs a relationship owner will ask, two from each kind of company the bet says should keep the records, and put only the company question to them. Before the first call, write the result that demotes the bet: if fewer than three of the eight confirm multi-year records of their own, the thesis stops driving outreach in this network. If CEOs confirm records that turn out to belong to their clients, rewrite the bet toward whoever owns them. Date every answer; a thesis resting on stale answers slides back into endorsement. For sharper questions to put to those eight CEOs, see [how to turn a contrarian data thesis into testable questions](/guides/how-to-turn-a-contrarian-data-thesis-into-testable-questions).

## Where attribution discipline breaks down

- **Attribution does not make the bet right.** Epoch AI's report on AI in 2030 argues scaling could continue for at least a few more years on existing public text and other kinds of data (our paraphrase), which cuts against any thesis built on a shortage of real-world records.
- **Rights.** A CEO can sincerely believe the records are the company's own while customer contracts say otherwise. Only SourceX's rights review and the company's counsel settle that.
- **Demand moves.** SourceX checks relevance to its current programs during review, and a company that misses one program may fit another later.
- **Tone travels.** The guide governs written and repeated claims. A relationship owner's enthusiasm on a call can still sound like an endorsement.
- **Over-hedging.** A message so qualified that nobody can act on it fails too.

**Methods note.** The four levels and the language guide were built on 2026-10-26 from SourceX's published screening criteria, the program terms (version 2026-10-22.2, since replaced by version 2026-11-04) and FTC staff guidance on endorsements. They classify statements; they do not measure how founders or LPs read them, which no published study covers. The eight-CEO threshold is an Illustrative default: replace it with your own, set before the first call.

## Next step

> **Refer a company. No cap on rewards.** Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

When a company question comes back with a dated yes from a CEO who wants the conversation, [register for a partner link](/register), check the facts against [who qualifies](/who-qualifies) and read the [program terms](/terms) before the relationship owner makes the introduction. More decisions like this one are collected in the [partner guides](/guides).

Deciding whether SourceX itself deserves capital is an investment question kept apart from any introduction; readers weighing it can request materials through the [SourceX investor overview](https://sourcex.si/investors).

This article is for information only and is not an offer to sell or a solicitation of an offer to buy securities. Any offering by SourceX is made only to verified accredited investors through its offering documents.

## Sources

- https://refer.sourcex.si/terms
- https://refer.sourcex.si/who-qualifies
- https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents
- https://www.ftc.gov/business-guidance/resources/ftcs-endorsement-guides-what-people-are-asking
- https://epoch.ai/files/AI_2030.pdf
- https://sourcex.si/investors
