A founder note is not a newsletter item. It reaches a handful of CEOs, it puts the fund's credibility behind every line, and it can be forwarded to people you never meant to address. The bar is therefore higher than for a market update: a founder should be able to verify each claim, see where the fund is guessing and say no at no cost to the relationship.
From category thesis to a question one CEO can answer
Category theses describe markets; founders run companies. The note's first job is translation. "Model developers will need more real records of how businesses operate" is a claim about an industry. The version a CEO can answer concerns their own archive: has the company kept, for several years, records it created itself of how work gets done, such as tickets, project histories, QA notes or dispatch exceptions, and do those records belong to the company rather than to its customers?
The same question decides who receives the note. Funding status plays no part in whether a company can be introduced. A company needs 50 or more full-time employees at peak, documented operating history, the rights to license and a CEO who wants the conversation; the CEO is also the person who authorizes any licensing discussion. A celebrated seed-stage company may fail all four tests, while a twelve-year-old portfolio company nobody mentions at the annual meeting may pass them. For the step from thesis to a named list of companies, the six-box activation canvas is the companion to this page; this page covers the document you send once the list exists.
The evidence-bounded founder memo
The memo has nine parts and fits on one page. Its limits are what make it credible.
| Part | What it says | Limit |
|---|---|---|
| 1. Date and shelf life | When it was written and when it will be rechecked | 90 days, then reread criteria and sources |
| 2. Thesis | One sentence, labeled as the fund's view | No market-size figure |
| 3. Evidence for | The strongest primary sources, each dated | Two at most |
| 4. Evidence against | The best case that the thesis is wrong | A full paragraph, never a footnote |
| 5. Your company's question | The thesis as a question about the CEO's own records | One question |
| 6. What we do not know | Demand for any company's records, rights, terms, outcome | Named plainly, not hedged away |
| 7. Our interest | That the fund may receive a referral reward from SourceX | Kept apart from anything about investing |
| 8. The ask | A reply for a short call, with no files | One action; silence means no |
| 9. What would change our mind | The replies or sources that would end the thesis | Written before sending |
Each sentence then gets the strongest verb its evidence supports, and no stronger.
| Evidence behind the sentence | Verb it allows | Never upgrade it to |
|---|---|---|
| Published analysis with a stated method | "estimates", "finds" | "proves", "will" |
| Company announcement | "says", "plans", "announced" | "is buying", "pays" |
| Program fact from the published terms | A plain statement, with the source | Your reading of what the terms intend |
| The fund's own inference | "we think", "our working view" | "the market is" |
| Nothing either way | "we do not know" | Silence |
| Private conversation or rumor | Leave it out | Anything at all |
micro1's announcement from October 2026 shows why the ladder matters. The company says it plans to spend $1 billion over the next 12 months acquiring and licensing enterprise operational data, with the stated aim of widening the industries and workflows in its reinforcement learning environments. A note may report that plan, with its date and its source. It may not say the money is being spent, that any portfolio company's records will be bought or that prices are rising, because the announcement shows none of those things.
Method note. The nine parts and the verb ladder are our editorial standard. They were built from SourceX's published program terms and screening profile (both checked 2026-10-26) and the sources cited on this page. They have not been tested on founders, and no reply rates for notes like this have been published; the five-founder test below is how a fund produces its own.
One memo, filled in (Illustrative)
Illustrative. An invented fund writing to five screened portfolio CEOs. The right-hand column shows the grade a reader would see if the evidence tags were printed.
| Part | Memo line (Illustrative) | Evidence tag |
|---|---|---|
| 1 | Written 26 October 2026; we will recheck it by 24 January 2027. | Shelf life |
| 2 | We think the records of how established companies do their work may become licensable for some of you. | Our view |
| 3 | Epoch AI's 2024 analysis estimates the stock of public human-generated text at about 300 trillion tokens and puts its full use by training runs between 2026 and 2032, at 80% confidence. | Published analysis, 2024 |
| 3 | micro1 says it plans to spend $1 billion over 12 months on enterprise data (announced October 2026); that is a plan, not a record of purchases. | Company announcement |
| 4 | Against us: a 2023 paper reports a 1.3-billion-parameter code model, trained partly on textbooks and exercises written by GPT-3.5, scoring 50.6% on HumanEval, and Epoch's own estimate has moved later before. | Published research, against |
| 5 | Has your company kept several years of its own tickets, project histories or QA notes, and are they yours rather than your customers'? | Program fact as a question |
| 6 | We do not know whether any program needs your records, what they would be worth or whether a license would follow; SourceX decides that in its review. | Unknown |
| 7 | If we introduce you and your company licenses data, our fund could receive a referral reward from SourceX, paid separately and never taken out of your payout. | Disclosure |
| 8 | Reply "yes" if you want a 20-minute call. Send no files. No reply means no, and we will not raise this again this year. | The ask |
| 9 | If three of you tell us your records mostly belong to customers, we will stop sending this note. | Test that ends it |
Notice what line 3 leaves out: it never says data is running out. Epoch's 2024 page explains that its estimate moved later than its earlier work had suggested, mainly because carefully filtered web data and repeated training passes enlarge the usable stock. A dated range from a group that has already revised itself belongs in the note as a range, with its date, and next to line 4.
Relationship and ownership evidence the note cannot replace
A well-graded note still fails if it reaches the wrong person or leans on information the fund should not use. Settle these facts for each recipient before anything is sent.
| Evidence | Where it should come from | Can the note use it? |
|---|---|---|
| The operating entity that holds the records | The CEO, or the company's own filings | Yes, by name |
| The current CEO, who authorizes | The relationship owner | Yes; the note is addressed to the CEO |
| Headcount at peak and years of records | A conversation with the CEO or relationship owner | Only as a question, never as an assertion |
| System names, volumes or record details from board packs | Board materials and information rights | No; that knowledge usually stays confidential |
| Who owns the relationship | Your own team | Internal only: that person sends or approves the note |
| Whether the records belong to customers | The CEO first, then SourceX's rights review | Only as a question |
Location is a published fact, not a judgment call. 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. What stays unknown until SourceX reviews a company: whether its records match a current program, whether customer contracts limit reuse and what a license would cover. Material that belongs to the company's clients, scraped data and any introduction the CEO has not been told about are usually not a fit, which is why line 5 asks about ownership before anything else. The founder explainer covers the conversation that follows a yes.
A five-founder test, and the replies that would change the thesis
Send the note to five CEOs whose companies pass every fit fact you can confirm, not to the portfolio mailing list. Allow 14 days, then count.
| Count after 14 days | What it tells you | What it changes |
|---|---|---|
| Yes replies | Whether the question lands | Zero from five: stop sending the note, then rewrite or drop the thesis |
| "Our records belong to customers" | Whether the rights premise holds in this portfolio | Three or more: the thesis is wrong for this fund |
| "Someone already raised this with us" | Whether others reached the company first, SourceX included | Record it; a company with earlier contact with SourceX is not new for attribution |
| Questions about money or timing | Whether parts 6 and 7 were clear | Rewrite those parts before any resend |
| Silence | A no | Do not chase; line 8 promised not to |
| Verified applications within 90 days | Whether interest turned into action | Report the count against its denominator of five |
The thesis can also change in the other direction. A primary source that revises the evidence for, a change in the published criteria, or a founder who describes records the note never anticipated each call for a rewritten, re-dated note. Whoever submits an introduction must first have signed the SourceX Channel Partner Agreement, and credit for a company goes to a single partner, the one whose valid introduction came first inside the 90-day window, and only once that company submits a verified application. For a larger, pre-registered version of this test, see the ten-introduction experiment; to keep a resent note from carrying stale criteria, use the versioned distribution process. A fund that later mentions the test to its LPs should report it as a method with its counts and no overstated results; value creation for emerging managers covers that kind of reporting.
What could make the note wrong
- Real records may not stay scarce. The phi-1 paper reports that, for code at least, model-written textbooks and exercises carried a small model to 50.6% on HumanEval. If that approach extends to business workflows, demand for licensed operating records could narrow.
- Forecasts get revised. Epoch's range has already moved later once. A note that states a window as fact ages badly; date every range.
- One buyer's plan is not a market. A single company announcement says nothing about prices, timing or which records any program wants.
- Rights fail quietly. SourceX's screening guidance treats material belonging to a company's clients as usually not a fit, and many software companies hold their most visible data inside customer accounts.
- Programs change. SourceX applies each program's own criteria, so a company that fits today may not next quarter, and the reverse.
- Founder attention is finite. Each note spends part of the relationship, and no published data shows how founders respond to notes like this. The founder fatigue audit weighs what else they are being asked.
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 screened CEO replies yes to your note, register for a partner link, hold the company up once more against who qualifies and reread the program terms your note relied on before you write the introduction. The portfolio side of the program is set out for venture platform teams, and more decisions sit in the guides hub.
Writing a founder note is referral work; judging whether SourceX itself merits an investment is a separate exercise, and a fund that wants to make that judgment can request investor materials via the SourceX investor overview.
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.