Deal economics
Dataset acceptance criteria diligence: a 10-point discretion score
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Dataset acceptance criteria diligence asks whether the buyer can refuse a conforming delivery. Score the clause on five elements: how objective the standard is, who runs which test on which version, whether the review window ends in deemed acceptance, whether rejection needs reasons and allows cure, and how long remedies last. Under IFRS 15, acceptance is a formality only when conformance can be objectively determined; otherwise a signed order behaves like an option.
Checklist
Dataset acceptance criteria
Key takeaways
- An acceptance clause decides whether a signed data order is earned revenue or an option the buyer holds.
- Revenue standards draw the same line: IFRS 15 treats acceptance as a formality only when the seller can objectively determine that a delivery meets the agreed specifications.
- Federal rules define acceptance as acknowledging conformance and, for prompt-payment interest, deem it on the 7th day after a conforming delivery unless there is a disagreement.
- The discretion score grades five clause elements from 0 to 2; contracts scoring 3 or less belong in pipeline, not backlog.
- Discretion and concentration compound, so ask for option-weighted concentration: option-band contract value held by the largest buyer.
What should dataset acceptance criteria diligence decide?
It should decide how much signed contract value a data company can treat as earned and how much is still an option the buyer holds. The acceptance clause makes that call, not the order form: a deal with a price, a dataset and a signature can still be refused at review if the clause lets the buyer turn down a conforming delivery.
Three features of the clause carry the decision. Objective standards determine whether conformance can be shown without the buyer's agreement. The review period determines how long the buyer can hold a delivery in limbo. Rejection mechanics determine whether a refusal must be explained, cured and limited to the defective part. Diligence that reads only price and term misses all three.
The evidence is documentary: the executed acceptance clause and its schedules, the test script or rubric it names, the dataset version it fixes, and the company's record of acceptances, rejections and cures by buyer. Acceptance-rate summaries become useful only after the clauses behind them have been read.
Why the wording of acceptance matters once revenue is counted
Because revenue standards tie timing to whether acceptance can be judged objectively. IFRS 15 says that if an entity can objectively determine that control of a good or service has been transferred in accordance with the agreed-upon specifications in the contract, customer acceptance is a formality. Its example is a clause based on size and weight, which a seller can check before the customer confirms, and it notes that experience with similar contracts can supply the evidence.
US GAAP asks the same question. ASU 2014-09 directs an entity weighing a contractual customer acceptance clause to paragraphs 606-10-55-85 through 55-88 when judging when control transfers. The practical reading for an investor: the more an acceptance test depends on the buyer's judgment, the weaker the case that a signed, delivered order is already earned, and the more reported backlog rests on the buyer's later decision.
Two companies with identical signed contract value can therefore carry very different risk. One has clauses a seller can satisfy by running a named test. The other has clauses that end with the words satisfactory to the buyer.
What public contract rules treat as a fair acceptance clause
Federal procurement rules make a useful benchmark because they are written for the buyer and still limit its discretion. FAR 46.501 defines acceptance as acknowledgment that supplies or services conform with applicable contract quality and quantity requirements, and lets it take place before, at or after delivery as the contract provides. The standard is conformance with written requirements, not satisfaction.
Silence has a limit too. For computing prompt-payment interest, FAR 32.904 deems government acceptance to occur constructively on the 7th day after the contractor delivers in accordance with the contract, unless there is a disagreement over quantity, quality or compliance. The same rule says constructive acceptance does not compel officials to accept, and a longer period, which is not available for commercial products and services, needs a documented justification and must not become routine agency practice.
Remedies are bounded as well. The commercial terms clause at FAR 52.212-4 lets the government require repair, replacement or reperformance of nonconforming items, or seek an equitable price reduction, and requires post-acceptance rights to be used within a reasonable time after the defect was or should have been discovered. Its alternate for time-and-materials and labor-hour contracts presumes services and materials accepted 60 days after delivery unless accepted earlier, and allows correction demands for up to 6 months after the last acceptance unless the contract sets another period.
Commercial law starts from a similar place for goods. Under UCC 2-606, a buyer accepts goods when, after a reasonable opportunity to inspect, it signals that they conform or that it will keep them anyway, when it fails to make an effective rejection, or when it acts as though it owns them. Article 2 governs sales of goods, and whether it reaches a data license is a question for counsel, so the contract's own clause usually carries the weight.
The acceptance discretion score: five elements, 0 to 2 points each
Score each material contract's acceptance clause on five elements, giving 2 points for terms a seller can satisfy on evidence and 0 for terms the buyer can decide alone. The total runs from 0 to 10. A score of 8 to 10 is earned: a conforming delivery will be accepted. A score of 4 to 7 is contingent: acceptance depends partly on judgment or timing. A score of 0 to 3 is an option: the buyer can, in effect, decide later whether it bought anything.
Score the executed clause with its schedules, never the template, and score a missing clause as 0 on every element. The rubric, with the evidence to request for each element:
| Element | 2 points: earned | 1 point: contingent | 0 points: option | Evidence to request |
|---|---|---|---|---|
| Objective standard | Measurable thresholds, or a rubric two qualified reviewers would apply the same way | Measurable thresholds plus judgment items under a written rubric | Satisfactory to the buyer, or acceptance at its sole discretion | Acceptance schedule, thresholds and rubric |
| Test and version | Named test or rubric and dataset version fixed at signing; either party can run it | Test named, but the buyer may revise it before review | No test named, or the buyer picks one after delivery | Test script or rubric, version identifiers, change log |
| Review window | Fixed window that ends in deemed acceptance unless a reasoned rejection arrives | Fixed window with no deemed acceptance | Open-ended review, or a start the buyer controls | Window terms, start event, delivery and notice records |
| Rejection and cure | Written reasons tied to a named criterion, a cure and redelivery right, and partial acceptance of conforming subsets | Reasons required, but no cure right or no partial acceptance | Rejection without stated reasons | Rejection notices and cure history by buyer |
| Remedy period | Post-acceptance claims time-limited, confined to defects the test could not reveal, with capped remedies | Time-limited claims with uncapped remedies | Open-ended revocation or refunds on demand | Warranty, refund and revocation clauses |
Three acceptance clauses scored (Illustrative)
Illustrative: three invented clauses show how the score separates contracts that a revenue line treats as equal. Clause A sets field-level completeness and schema thresholds in a schedule, names the test script and dataset version, closes a fixed window with deemed acceptance, requires reasons, allows a cure and partial acceptance, and limits later claims in time but not in amount. It scores 9 and is earned.
Clause B uses the same thresholds but adds a fitness test for the buyer's intended training use, judged by the buyer, lets the buyer update the test before review, keeps a fixed window with deemed acceptance, requires reasons without a cure right and allows a refund on demand. It scores 5 and is contingent. Clause C makes acceptance subject to the buyer's sole discretion, names no test, starts review whenever the buyer begins its evaluation, requires written reasons with no cure, and limits later claims in time. It scores 2: an option.
Counted the usual way, all three are signed contracts of equal standing. Scored, only Clause A belongs in backlog without a discount. The figure sets the three scores side by side, and the element-by-element points are:
| Element | Clause A | Clause B | Clause C |
|---|---|---|---|
| Objective standard | 2 | 1 | 0 |
| Test and version | 2 | 1 | 0 |
| Review window | 2 | 2 | 0 |
| Rejection and cure | 2 | 1 | 1 |
| Remedy period | 1 | 0 | 1 |
| Total and band | 9, earned | 5, contingent | 2, option |
Acceptance discretion against buyer concentration: where signed value is weakest
Discretion and concentration compound, so score them together. A discretionary clause with a small buyer is a nuisance; the same clause with the buyer that supplies most of the revenue is leverage over price, scope and timing. The figure to ask for is option-weighted concentration: the share of total contracted value that is both in the option band and held by the largest buyer.
Large buyers often contract on their own paper, which is how a prestigious name or a large mandate can create dependence without any obligation to accept or to buy again. Check whose template each acceptance clause came from, and whether the company has ever relied on a deemed acceptance against its largest buyer. Licensing businesses are not immune: Reddit's FY2025 10-K says "substantially all of the contract value associated with our licensing revenue is derived from two of our partners." The buyer concentration and commitment matrix for acceptance:
| Discretion band | Spread across many buyers | Held by the largest buyer | How to count it |
|---|---|---|---|
| Earned (8 to 10) | Low acceptance risk; acceptance follows conformance | Concentration risk remains, but each delivery can be earned on evidence | Count as backlog; stress-test renewal, not acceptance |
| Contingent (4 to 7) | Timing risk; expect some deferrals | The buyer can slow revenue and reopen scope at each review | Discount by the company's own rejection and cure history |
| Option (0 to 3) | Pipeline, not backlog | The buyer decides later whether it bought; the highest risk cell | Exclude from backlog until acceptance is recorded |
| Clause not produced | Treat as option until the executed text is seen | Treat as option and ask why it is missing | Exclude, and make production a condition of diligence |
Is a demanding buyer a red flag? The case for strict criteria
No. Strictness and discretion are different things, and the score rewards strict criteria that are objective. A clause with a high completeness threshold, a residual personal data limit and a tight schema check can score 10, because a seller can show conformance on evidence. What the score penalizes is a bar nobody can locate in advance.
Buyers have real reasons to want judgment in acceptance, because training value is hard to specify. Anthropic's guidance on agent evaluations shows how judgment can still be made objective: "A good task is one where two domain experts would independently reach the same pass/fail verdict." The same guidance warns that ambiguity in task specifications becomes noise in metrics, and that everything a grader checks should be clear from the task description. A rubric that meets that test belongs in the 2-point column even when people apply it.
Buyer exclusions work the same way. OpenAI's data partnerships page says it is not seeking data that contains sensitive or personal information or that belongs to a third party. Written as a measurable residual-data limit and a documented rights check, an exclusion like that is an objective acceptance criterion. Left as a general reservation, it is discretion.
What could make the discretion score misleading
Five limits apply, and each is a data procurement risk to check before relying on a score:
- Paper is not behavior. A buyer with an earned-band clause can still stop ordering after a dispute, so pair the score with renewal and reorder history.
- Objective tests can be wrong. A delivery can pass every threshold and still disappoint in training. Deemed acceptance closes a window; it does not prove the data was useful, and FAR 32.904 itself says constructive acceptance does not compel officials to accept.
- Selection bias. Companies show their best clauses first. Score every contract above a stated size and record which ones were not produced.
- Enforceability. Whether a deemed-acceptance or limitation clause holds depends on the governing law and the facts, which counsel must review; UCC Article 2 defaults apply to goods and may not reach a data license.
- Concentration overrides the score. When two partners hold substantially all licensing contract value, as Reddit's filing describes, even earned-band clauses leave a company exposed at renewal.
Acceptance questions for any data company, SourceX included
SourceX works with companies licensing selected business records to AI developers, and its role runs from sourcing and qualifying data and reviewing rights to contracting, preparing and delivering the data, and handling payment; delivery waits for the company's signature on a scope and terms it has approved. Those facts show where acceptance sits in the process; they say nothing about acceptance outcomes. SourceX has not published acceptance rates, rejection histories or how its contracts would score on discretion, and it shares company and commercial metrics with investors directly.
Its capital raise is open only to accredited investors under Rule 506(c), and anyone evaluating the company can request investor materials on its investor page. The questions below apply to SourceX and to any other data company:
- What is the discretion score of every contract above a stated size, and who scored it?
- What share of contracted value sits in the option band, and how much of that is held by the largest buyer?
- Which test script, rubric and dataset version does each clause name, and who may change them?
- How many deliveries were rejected, cured and then accepted, by buyer and by reason?
- Has the company ever relied on a deemed acceptance, and with which buyer?
- Which clauses came from the buyer's template, and what changed in negotiation?
Methods note: how the discretion score was built and what it cannot see
The five-element score, its bands, the option-weighted concentration figure and the three clauses are editorial constructs built for this article on 26 October 2026. They draw on IFRS 15's acceptance guidance, the ASU 2014-09 cross-reference, FAR 46.501, 32.904 and 52.212-4 read from GSA's published FAR source files, UCC 2-606, Anthropic's evaluation guidance, OpenAI's data partnerships page and Reddit's FY2025 10-K.
Not measured: no executed contracts were scored, no rejection histories were collected and no public company's acceptance clauses were coded. Clauses A, B and C are illustrative. The five elements are weighted equally by design; a reader who thinks the standard matters more than the remedy period can double its points and rescore. No SourceX data is used.
Frequently asked questions
What are dataset acceptance criteria?
They are the written tests a delivered dataset must pass before the buyer is treated as having accepted it, such as completeness against source counts, schema conformance, label accuracy on a sample or a residual personal data limit. Good criteria name the test, the threshold and the dataset version, so either party can tell in advance whether a delivery conforms.
Does deemed acceptance prove a dataset is good?
No. Deemed acceptance closes a review window when the buyer does not reject in time; it says nothing about whether the data performed. Even the federal prompt-payment rule that deems acceptance on the 7th day for interest purposes states that it does not compel officials to accept.
Why should an investor read acceptance clauses?
Because they decide whether signed contract value is earned or still optional for the buyer. Under IFRS 15, acceptance is a formality only when the seller can objectively determine conformance with the agreed specifications, so subjective clauses move revenue timing toward the buyer's decision.
What makes an acceptance criterion objective?
A measurable threshold, a named test and a fixed dataset version, applied so that two qualified reviewers would reach the same verdict. A judgment-based rubric can qualify if it is written clearly enough to pass that two-reviewer test.
Does SourceX publish acceptance rates?
No. SourceX has not published acceptance rates or how its contracts score on acceptance discretion. It shares company and commercial metrics with investors directly, and materials are available on request through sourcex.si/investors.
Sources
- IFRS 15 says that if an entity can objectively determine that control has transferred in accordance with agreed-upon specifications, customer acceptance is a formality; experience with similar contracts can provide evidence. Source
- ASU 2014-09 directs entities evaluating a contractual customer acceptance clause to paragraphs 606-10-55-85 through 55-88. Source
- FAR 46.501 defines acceptance as acknowledgment that supplies or services conform with contract quality and quantity requirements, taking place before, at or after delivery. Source
- FAR 32.904 deems acceptance constructively on the 7th day after conforming delivery for prompt-payment interest, unless there is a disagreement; it does not compel acceptance, and longer periods need documented justification. Source
- FAR 52.212-4 allows repair, replacement, reperformance or an equitable price reduction and limits post-acceptance rights to a reasonable time; its Alternate I presumes acceptance 60 days after delivery and allows correction for up to 6 months. Source
- UCC 2-606 says a buyer accepts goods by signaling conformity after a reasonable opportunity to inspect, by failing to make an effective rejection, or by acting inconsistently with the seller's ownership. Source
- Anthropic's evaluation guidance says a good task is one where two domain experts would independently reach the same pass/fail verdict and that ambiguity in task specifications becomes noise in metrics. Source
- OpenAI's data partnerships page says it is not seeking data with sensitive or personal information or data belonging to a third party. Source
- Reddit's FY2025 10-K says substantially all of the contract value associated with its licensing revenue is derived from two of its partners. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.