Getting started
Supplier data eligibility is not an offer: the four gates in between
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
The supplier data eligibility-offer distinction is about who has decided what. Eligibility means a company and dataset pass published screening rules. An offer means a buyer with a matching task has put definite terms in writing. Accepted terms mean both sides have signed. Each gate has its own owner and evidence, so passing the first predicts little about the rest. Spend little beyond screening until a definite written offer exists.
Process
Supplier data eligibility
Key takeaways
- Eligibility, a matched buyer request, a written offer and signed terms are four separate gates with four different owners.
- An offer is addressed to you, definite, and binding once accepted; a screening result or an estimate range shown to every applicant is not.
- Demand is narrow and concentrated: buyer exclusions and a short list of buyers filter out many datasets that pass screening.
- Diagnose a no-offer outcome before acting; only one of six common reasons is fixed by preparing more data.
- Cap effort at the gate already passed, and reconcile any offer that arrives line by line against the scope that was screened.
What is the supplier data eligibility-offer distinction?
Eligibility is a screening result about the company and the dataset; an offer is a buyer's definite proposal of terms for specific records. The first is judged against rules that apply to every applicant, the second against one buyer's task, budget and schedule, so a dataset can pass the first and never reach the second.
Contract law has drawn a version of this line for decades. Article 14 of the UN Convention on Contracts for the International Sale of Goods treats a proposal as an offer only when it is addressed to one or more specific persons, is sufficiently definite and shows an intention to be bound if accepted; a proposal made to the public generally counts only as an invitation to make offers. The convention governs international sales of goods, not data licenses, but its test is a clean way to read anything a supplier receives. An eligibility notice, an estimate range or a program page written for all comers fails at least one of the three conditions.
Between eligibility and an offer sits a step suppliers rarely see: a buyer request that matches the dataset. After the offer sits another: terms both sides accept and sign. That makes four gates, and each one is decided by someone different.
Four gates, four owners: who decides each one and on what evidence
Each gate answers its own question and leaves the others open, so evidence for one cannot be borrowed for the next. A positive screening result is strong evidence for gate one and no evidence at all for gates two to four.
The column leaders most often skip is the last one, which says what passing a gate does not tell you:
| Gate | Question it answers | Who decides | Evidence that it is passed | What it does not tell you |
|---|---|---|---|---|
| 1. Eligible | Do the company and dataset meet the screening rules? | The screener, applying rules that cover every applicant | A dated assessment result that names the dataset | Whether any buyer wants these records |
| 2. Matched | Does a buyer's request fit this dataset's scope? | The buyer, through its stated scope, volume, language, format and timing | A written request naming the record type and period | Price, conditions, or whether rights review will pass |
| 3. Offered | Has someone proposed definite terms for these records? | The buyer, or the intermediary acting on the request | A written offer stating scope, uses, conditions and payment triggers | Whether the terms survive review and get signed |
| 4. Accepted | Have both sides agreed to the same terms? | Both parties, through authorized signatories | A signed agreement and any deal-specific schedule | Whether the delivery will pass the recipient's review |
Why eligible supply outnumbers matched demand
Datasets that pass screening usually outnumber buyer requests because the two are produced differently. Screening rules are fixed and applied to everyone who applies, while demand arrives one task at a time from a small number of developers, each with its own exclusions.
Exclusions do much of the filtering. OpenAI's data partnerships page describes interest in large-scale datasets that are not already easily accessible online and that express human intention, and it says what it does not want: data carrying sensitive or personal information, or information that belongs to someone other than the party offering it. A dataset can clear every screening rule and still fall outside one buyer's request, because the part that makes it valuable is the part that buyer has chosen not to take.
Concentration does the rest. Reddit, whose data licensing deals are among the most widely reported, states in its Form 10-Q for the first quarter of 2026 that "to date, substantially all of the contract value associated with our licensing revenue is derived from two of our partners". If a large public licensor's licensing revenue rests on two counterparties, a mid-size company with an eligible archive should assume that only a few buyers will ever look at it, and that none may be looking this quarter.
Six reasons an eligible dataset receives no offer
Most no-offer outcomes fit one of six patterns, and each calls for a different response. Diagnose before acting, because the usual reflex, preparing more data, fixes only one of them.
Rights questions stall more datasets than suppliers expect, because buyers have learned not to trust labels. Licensing information was missing for more than 70 percent of the datasets on widely used hosting sites, with error rates above 50 percent, according to an MIT Media Lab audit that examined more than 1,800 text datasets and appeared in Nature Machine Intelligence in 2024, so a buyer asks for the contract rather than the description.
Where an intermediary is involved, ask for the reason in writing, and record your own evidence for it either way:
| Reason | What it looks like | Evidence to ask for | Supplier response |
|---|---|---|---|
| No matching request | Silence after a positive screening result | Whether any open request names this record type or period | Hold, set a review date and spend nothing more |
| Buyer exclusion | Interest fades once content details are shared | Which categories the buyer excludes, such as personal or third-party content | Narrow to records outside the exclusion, or close the file |
| Scope mismatch | Requests exist but name other years, languages or formats | The gap between requested and screened scope, line by line | Resubmit only if records you already hold close the gap |
| Rights unproven | Questions about contracts or notices stay unanswered | Which documents the reviewer still needs | Have counsel read a contract sample before any other work |
| Preparation outweighs value | The request implies cleanup larger than it justifies | The preparation steps requested and who would carry them out | Price the effort; decline if it cannot be narrowed or shared |
| Timing | A request was filled by another supplier or paused | Whether the request is closed, paused or expected to reopen | Keep the scope note current so a new request can be matched fast |
Illustrative: two eligible record families, one request
Illustrative only; the company and figures are invented. A 400-person regional HVAC and plumbing contractor puts two record families through screening: eight years of closed work orders with technician notes, and four years of recorded customer calls. Both pass. The company meets the size and country rules, the records are in English, and each family holds more than 100,000 records in a single system.
Over the following quarter, one request arrives for field-service records that show how technicians diagnose and resolve faults. It matches the work orders on record type and period. Nothing arrives for the calls. When the company asks why, the answer fits two of the six reasons at once: the open requests exclude voice data containing personal information, and no request names call recordings at all.
The memo that follows treats the two families differently. The work orders move to gate two: counsel reads a sample of customer contracts, the system owner answers the buyer's scope questions, and nothing else starts until a written offer exists. The call recordings stay at gate one with a review date and no further spending. Had the company prepared both families at the screening stage, every hour spent on the recordings would now sit unused, and the prepared copies would carry privacy duties for no purpose.
How much effort each gate justifies
Match effort to the gate already passed, not to the gate you hope to reach. Work done before an offer is lost entirely if none arrives; work done after signed terms is the work the deal pays for.
Accountants draw the same line. Under paragraph 9 of IFRS 15, an arrangement is treated as a contract only once the parties have approved it, each party's rights and the payment terms can be identified, it has commercial substance and collection is probable. Neither a screening result nor an unanswered request meets that test, which is why neither belongs in a budget as income.
Public procurement offers a parallel. The Federal Acquisition Regulation, at 9.103, bars any federal purchase or award unless the contracting officer first makes an affirmative determination that the prospective contractor is responsible. That finding is a precondition for an award, not the award itself, and eligibility plays the same role in data licensing. Use the worksheet to set a ceiling for each stage:
| Gate reached | Work justified now | Work to defer | Stop signal |
|---|---|---|---|
| Before screening | Answer screening questions from system knowledge, with no exports | Exports, samples and any cleanup | The company or dataset fails a published rule |
| Eligible, no request | Keep a one-page scope note current: systems, years, counts, known limits | Contract review beyond a small sample; all preparation | No matching request by your own review date |
| Request matched, no offer | Answer the buyer's scope questions; counsel reads a contract sample | De-identification, full exports and delivery planning | A buyer exclusion covers the core records |
| Written offer received | Reconcile the offer to the screened scope; estimate effort by role | Delivery work until terms are signed | Conditions or triggers the company cannot meet |
| Terms signed | Preparation, de-identification and delivery under the agreement | Nothing; this is the work the deal pays for | A rights or privacy finding the agreement does not cover |
When an offer arrives, reconcile it against what was screened
Compare the written offer with the scope that passed screening, line by line, before treating it as good news. Offers drift: a buyer may want fewer systems, other years, more fields or stricter conditions than the assessment assumed, and the effort estimate has to follow the records actually requested rather than the ones first described.
Write every difference down, even small ones, because a sequence of small narrowings can leave a much thinner deal than the one leadership approved. Four checks catch most drift:
- Records: the systems, record types, date range and fields named in the offer against those in the assessment, with each difference noted.
- Uses: what the recipient may do with the data, for how long, whether it may pass the data on, and whether any exclusivity binds the company.
- Conditions: the steps the company must complete, such as a de-identification standard or a delivery format, and who carries out each one.
- Payment triggers: the exact events that make an amount payable, quoted from the offer, and any review that sits between delivery and payment.
Isn't screening pointless if it predicts so little?
No, but its value lies in what it rules out, not in what it promises. Screening is cheap, needs no records and removes the clear no's early. It also leaves a written scope that lets a later request be matched quickly. What it cannot do is stand in for demand.
The opposite mistake is common in planning. A positive screening result appears in a budget, a board update or a lender conversation as if it were a pipeline entry. Eligibility and indicative ranges never promise selection, a closing or a particular amount, and a figure attached to them anchors expectations that a later, narrower offer then disappoints.
A fair test of whether screening was worth doing is whether, a year later, the company knows more than before: which records it holds, which rights questions are open and what a buyer would need. If so, the screening paid for itself even without an offer.
The decision memo for an eligible dataset still waiting
Record the status in a short memo so the waiting period has rules. The memo prevents two opposite failures: sliding into preparation work nobody authorized, and forgetting the dataset until a request arrives that the company can no longer answer quickly.
At a glance
The decision memo for an eligible dataset still waiting
Its limits belong on the page. It covers one dataset on one date, reports no study result, and is neither legal advice nor a valuation. The next evidence it needs is a written request, a contract sample read by counsel, or a change in the company's systems or history. The fields:
- Gate reached and its date, with the document that shows it.
- Scope as screened: systems, record types, years, approximate counts and known exclusions.
- Demand evidence: none, described verbally, or a written request, each with dates.
- Diagnosis: which of the six reasons applies, and the evidence for it.
- Effort spent so far and the ceiling until the next gate, by role.
- Reopen triggers: a matching written request, a resolved rights question, a new system or a longer history.
- Decision: hold, narrow and resubmit, fix one named gap, or close the file.
- Review date and the person who owns it.
How eligibility, requests and signed terms work at SourceX
SourceX keeps the gates apart in the same way. Its first step is a short company assessment that needs no data uploads. It works with U.S., Canadian and EU companies that had at least 50 full-time employees at their peak, and companies in several other markets can be considered when at least 90% of the offered records are in English. For those markets eligibility turns on the dataset rather than the company's country, and every dataset still goes through rights, privacy, quality and buyer-fit review.
Demand comes from buyers' requests. SourceX sources datasets for AI buyers on request rather than holding them in stock, and buyers set out scope, volume, language, format, timing and licensing needs. As a rough yardstick, buyers look for one system holding tens of thousands of records, with history kept consistent for typically two years or more; smaller sets can qualify when the work they capture is rare, and each case is assessed on its own facts.
Acceptance is explicit and comes first: the company signs off on scope and terms and puts its name to the agreement, and only then delivers any data, with exclusivity and term fixed in each deal's signed agreement. A positive assessment is therefore gate one. It is not an offer, and it does not promise a payout.
Frequently asked questions
Does passing an eligibility check mean a buyer wants my data?
No. Eligibility says the company and dataset meet screening rules; demand exists only when a buyer's request names records like yours. Treat a positive result as a reason to keep a short scope note current, not as a sign that an offer is coming.
Is an estimate range or indicative figure the same as an offer?
No. An offer is addressed to you, states definite terms for specific records and shows an intention to be bound if you accept. A range shown to every applicant, or a figure that depends on later review, meets none of those conditions.
How long should a company wait for an offer after qualifying?
There is no standard period, so set your own. Pick a review date, spend nothing beyond a scope note until then, and at the review either hold, narrow the scope, fix one named gap or close the file.
What can a supplier do to make an offer more likely?
Fix what is within its control: describe the scope precisely, clear rights questions with a contract sample read by counsel, and exclude record types buyers commonly refuse, such as third-party content. It cannot create demand, so effort beyond that should wait for a matching request.
Should an eligible dataset appear in a forecast or board update?
As a fact, yes; as income, no. Report it as eligible with no request and the date, and keep any amount out until a written offer exists. Accounting standards such as IFRS 15 do not treat an arrangement as a contract before approved terms exist.
Sources
- Article 14 of the UN Convention on Contracts for the International Sale of Goods treats a proposal as an offer only when it is addressed to specific persons, sufficiently definite and shows an intention to be bound; a proposal to the public is generally an invitation to make offers. Source
- OpenAI's data partnerships page seeks large-scale data not easily accessible online that expresses human intention, and says it is not seeking sensitive or personal information or information belonging to a third party. Source
- Reddit's Form 10-Q for the first quarter of 2026 states that to date substantially all of the contract value associated with its licensing revenue is derived from two of its partners. Source
- The MIT Media Lab audit of more than 1,800 text datasets (Nature Machine Intelligence, 2024) found license omission above 70 percent and error rates above 50 percent on widely used hosting sites. Source
- IFRS 15 paragraph 9 treats an arrangement as a contract only once it is approved, rights and payment terms are identifiable, it has commercial substance and collection is probable. Source
- FAR 9.103 bars any purchase or award unless the contracting officer makes an affirmative determination that the prospective contractor is responsible. Source
See if your company qualifies
A short company assessment. No data uploads are needed.