# Why not every referred record is usable | SourceX

Canonical: https://refer.sourcex.si/guides/why-a-referral-does-not-mean-every-record-is-usable
Last updated: 2026-11-04

Because suitability is decided record family by record family, after the introduction. Material that belongs to clients, scraped or personal consumer data, records already licensed for AI training, records altered with AI for the sale, and files restricted by contracts can all fall outside on rights. Personal details are removed, which can take fields with them, and what remains is judged on completeness and consistency. Tell the founder first: a referral opens a review, and the usable share is unknown until it ends.

That sentence protects the founder and you. A founder who expects every record to count will read a narrower result as rejection, or as a partner's overselling. A founder who has heard the boundaries in advance reads the same result as the review working as described.

## What a founder hears when you say "not every record"

Said carelessly, "not every record will be usable" sounds like "your data is probably junk". Founders tend to hear one of four things in it.

- **A judgment on the company.** As if years of work had been graded and failed.
- **Wasted effort.** A fear of spending weeks on exports for records that are later thrown out.
- **A hidden catch.** That something will surface late in the process and sink everything.
- **An overselling partner.** That you knew all along and talked it up anyway.

The bounded answer to all four is the same: the boundaries are published, most of them can be discussed at metadata level before any file moves, and falling outside one is a statement about rights, privacy or record shape, not about the business.

## Seven suitability boundaries

Each boundary comes from a SourceX page. The last column says who settles it; no row is settled by the partner.

| Boundary | What falls outside it | What SourceX publishes | Who settles it |
|---|---|---|---|
| 1. Whose records | Material that belongs to the company's clients rather than the company | Listed as usually not a fit on SourceX's screening page | The company's counsel and SourceX's rights review |
| 2. Origin and control | Scraped data, personal consumer data, data the company does not control | Also listed as usually not a fit | SourceX's rights review |
| 3. Contracts and confidentiality | Records the company cannot license without breaching client confidentiality, employee privacy or third-party contracts | The screening page's definition of rights to license | Counsel reading the contracts, then SourceX |
| 4. An earlier AI-training license | Data already licensed for AI training | The supplier confirms the data has not been licensed for AI training before | The CEO, who confirms it |
| 5. Authenticity | Data created or altered with AI for the sale | The supplier confirms this too | The CEO, who confirms it |
| 6. Personal details | Names, contact details and other identifiers, and sometimes fields that carry them | Personal details are removed before sharing, through approved processing routes | The processing route set in the company's Confirmation |
| 7. Shape and consistency | Records without an outcome, fields that changed meaning, histories that do not connect | SourceX's guide: usable records are complete, consistent and connected; quality affects value | SourceX's review and preparation |

**Boundary 6 deserves an honest sentence.** De-identification is not free. NIST's 2023 guidance on de-identifying datasets, SP 800-188, describes a trade-off between identifiability and accuracy, with accuracy falling as more aggressive techniques are used, while noting that stronger protection does not always reduce usefulness for a specified purpose (a paraphrase of the search excerpt). For a founder, that means free-text records dense with customer names may come through with gaps.

**Boundary 7 cuts both ways.** SourceX's guide on usable records says records need not be perfect and that preparation fixes a lot, but that gaps and messy fields affect value. A boundary here is usually a question of which years or systems hold together, not a yes-or-no on the company.

**Demand is a different cut.** Records that pass all seven boundaries may still not fit a current program, or only part of them may. That is the subject of [explaining partial dataset interest](/guides/how-to-explain-partial-dataset-interest-at-referral-stage); this page covers the cuts that happen whatever the demand.

## One inventory through the boundaries (Illustrative)

Illustrative. A generic 210-person residential property manager with nine invented record families; no outcome is implied.

| Record family | Boundary it meets | Where it lands |
|---|---|---|
| Maintenance work orders, 2018 to 2025, with closure codes | None found at metadata level | Goes to review |
| Maintenance work orders, 2012 to 2017, from an older system without closure codes | 7: no recorded outcome, fields recorded differently | Out for now; may be reconsidered if outcomes can be linked |
| Tenant service emails | 6: dense with names and unit numbers | Goes to review, with identifiers removed and some fields lost |
| Owner statements that the management contracts assign to each property owner | 1: the clients' material | Out |
| Competitor rental listings collected from websites | 2: scraped | Out |
| Vendor bids received under contracts that bar disclosure | 3: third-party contract terms | Out unless counsel finds otherwise |
| Unit inspection notes licensed to an AI vendor two years earlier | 4: already licensed for AI training | Out |
| A tidied export of call summaries rewritten with an AI tool before the review | 5: altered with AI for the sale | Out; the original summaries could be considered instead |
| Staff-written procedures and training manuals | None found at metadata level | Goes to review |

Nine families listed, three go to review. Every exclusion in the table was visible from a conversation with the CEO about what each family is, who created it and what has happened to it. No file was opened.

Two things are worth noticing. The excluded families include some of the largest by volume, so a big archive is no guide to a big usable share. And two of the exits were avoidable: the AI-rewritten export would have been fine as originals, and the older work orders might return if their outcomes can be linked. Ask about both before the company prepares anything.

## Fixed rules, case-by-case calls and the unreviewed remainder

**Published, and quick to test:** boundaries 1, 2, 4 and 5. SourceX lists client-owned material and scraped or personal consumer data as usually not a fit, and asks every supplier to confirm that its data was never licensed for AI training and was not created or altered with AI for the sale. A CEO can usually answer the questions behind all four in one conversation.

**Case by case:** boundaries 3, 6 and 7. Whether a contract permits licensing, how much a record loses when personal details are removed and which years hold together are judged on the actual records and contracts, by counsel and in SourceX's review.

**Unreviewed at referral stage:** whether what passes fits a current program, how much of it, and on what terms. Only what reaches Closing is paid, and a partner reward is calculated on that payment alone, never on records that were listed and set aside. For a fuller list of factors that lower value, see [what reduces data value](/resources/what-reduces-data-value-for-ai-training); for sensitive categories, the [sensitive data question](/questions/is-my-company-data-too-sensitive-to-license).

## A bounded answer for the skeptical founder, and where it goes next

Founders often test the point with one sharp question. Three common versions, with answers that stay inside what is known:

- **"So you think most of our data is useless?"** "No. I don't know yet, and neither does anyone else. SourceX publishes the boundaries; most of them we can check together from a description, without opening a file."
- **"Why introduce us at all, then?"** "Because the families that pass, like your recent work orders and your procedures, are the kind of company-created records SourceX screens for. Whether they fit a current program is the review's job."
- **"What if something sinks it late?"** "The boundaries that can surprise you are the contract and privacy ones. Those go to your counsel and SourceX's rights review early, before anyone prepares an export."

The escalation path follows the boundary. Contract and ownership questions go to the company's counsel and SourceX's rights review. Questions about earlier AI licenses or AI-altered material are confirmed by the CEO. De-identification is settled in the processing route the Confirmation sets. The partner relays none of these answers and interprets none of them.

## One voluntary step, and an easy no

Offer one step: the CEO runs the nine-row exercise above on the company's own record families, using SourceX's short assessment, which asks for no files, and notes any family that meets boundary 1, 2, 4 or 5. If the CEO wants to fence off families before anyone looks, the [scope boundary approach](/guides/when-the-company-wants-to-limit-which-records-are-discussed) does that from the company's side.

Make declining easy. If the exercise leaves nothing the CEO is willing to discuss, the answer is a dated "no fit for now", and nothing is lost.

The company must pass the size and authority screen as well: 50 or more full-time employees at peak (contractors excluded), documented operating history, rights to license its records and a CEO who wants 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.

## Where boundary illustrations mislead

- **Real records are messier.** One family can straddle two boundaries; a ticket system can hold the company's own notes and a client's attachments side by side. The review splits them, not the partner.
- **Contracts surprise people.** Boundary 3 depends on wording a CEO may not remember. Treat any "we own all of it" answer as a question for counsel.
- **De-identification results vary.** How much a record loses depends on its content and the processing route, which this page cannot predict.
- **No published usable-share figures.** SourceX has not published what share of listed records typically passes review, and a partner should never estimate one.
- **For investors, boundaries are process, not performance.** SourceX's investor overview says ownership checks and removal of personal details are built into its workflow; it shares company and commercial metrics with investors directly, and nothing here is such a metric.

**Method note.** The seven boundaries were compiled on 2026-10-27 from SourceX's screening page, its process page and its guide to usable records, with NIST SP 800-188 for the de-identification trade-off. The nine-family inventory is invented to show how the boundaries interact; it does not measure how often any boundary applies. To test the page on real conversations, log each introduction's listed families, the boundary each one met at metadata level and the families that went to review, and compare that with what review later found.

## 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 the founder has heard the boundaries and at least one record family still looks like the company's own, [join as a partner](/register), match the company to [who qualifies](/who-qualifies) and see in the [terms](/terms) why only closed, paid data counts toward a reward.

Explaining suitability boundaries is part of a careful introduction; evaluating SourceX as an investment is a separate matter, and investors can ask for 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/who-qualifies
- https://sourcex.si/how-it-works
- https://sourcex.si/guides/business-data-quality-for-licensing
- https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-188.pdf
- https://refer.sourcex.si/terms
- https://sourcex.si/investors
