Skip to content

Deal economics

Data rights defensibility: five links, and the tests that break them

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Data rights create defensibility only when five links hold at once: the right is valid, enforceable against substitutes, necessary to the buyer's result, renewing with new records, and scarce across other holders. Score each link from 0 to 2 and let the weakest set the verdict, because one broken link turns a moat into a contract. Exclusivity can strengthen a chain; it cannot repair a broken one.

License terms

Data rights defensibility

This page covers data rights defensibility: five links, and the tests that break them.Illustrative

Key takeaways

  • A data right is a legal permission; defensibility is the economic result of repeat demand that rivals cannot meet another way.
  • Score five links, valid, enforceable, necessary, renewing and scarce, and use the lowest score as the verdict rather than the total.
  • Flow rights to records created after signing support repeat demand; archive rights are usually sold once per buyer.
  • Enforceability differs by jurisdiction: the 2025 Ross ruling enforced rights over curated headnotes, while EU law permits mining of public content without a valid reservation.
  • Exclusivity raises scarcity for its term but cannot repair missing authority or a buyer's ability to use substitutes.

When do data rights create defensibility?

Data rights create defensibility when they produce repeat demand that a competitor cannot meet another way. A right is a legal permission; defensibility is an economic result. Five links connect the two, valid, enforceable, necessary, renewing and scarce, and the chain is only as strong as its weakest link.

Investors usually meet the claim in compressed form: we hold the rights, so we have a moat. The compressed version skips every question that decides value. Is the right real, and can its holder stop others from using the same material? Does the buyer's result actually depend on it? Does it cover records that keep arriving, or only an archive a buyer consumes once? Could the buyer get comparable rights from someone else?

This page turns those questions into a scorecard with an explicit disconfirmation test for each link, scores three illustrative rights positions, sets out an evidence design that could prove a rights-based moat claim wrong, and names the forces that erode one after it is scored. It looks at the rights themselves; whether a dataset outperforms its substitutes is one input to the necessity link, not the whole test.

Stock rights and flow rights earn differently

A right to a fixed archive is usually sold once per buyer, while a right to records that keep arriving can be sold again each period. Repeat demand, the economic core of defensibility, lives mostly in flow rights.

Once a model has trained on an archive, the buyer keeps what the model learned, and a second purchase of the same archive adds little. A flow right covers records created after signing, so each period brings material the buyer has not yet seen, and the right is renewed because the buyer's need renews.

The public deals that disclose their structure point the same way. OpenAI's announcement of a multi-year global partnership with News Corp gives OpenAI access to current and archived content from News Corp publications, pairing a stock with a flow. Reddit's S-1 describes data licensing arrangements entered in January 2024 with terms of two to three years: time-limited access rather than a one-time sale.

Flow rights also expose the test investors care about most, which is renewal. Reddit's annual report for 2024 states that "substantially all of the contract value associated with our licensing revenue is derived from two of our partners", and it warns that those partners could let the arrangements lapse or renew them on less favorable terms. A flow right creates the chance of repeat demand; only renewals on the record prove it.

Score each of five links 0, 1 or 2 and let the lowest score set the verdict. The total is not the answer: a right that scores 2 on four links and 0 on one is a contract, not a moat, because a single broken link lets demand or supply route around it.

Each link has one test designed to break it. Run the tests in order, because a validity failure makes the rest moot, and record the evidence behind every score so another reviewer can rescore it. A 1 means the link holds with a named weakness; that weakness belongs in the investment memo beside the score.

The five-link rights scorecard, with a test that breaks each link
LinkScores 2 whenScores 0 whenDisconfirmation test
ValidEvery sampled record traces to a signed grant from someone with authority to make itAny sampled record traces to no grant, or the grantor lacked authorityDraw records at random and ask for the grant and signer behind each one
EnforceableThe holder can stop unlicensed use through contract, access control or lawRivals can lawfully reach the same material without a licenseAsk counsel where a competitor could legally obtain the same content without paying
NecessaryThe buyer's acceptance test fails without the licensed dataPublic, synthetic or in-house data passes the same testRerun the buyer's test with the best substitute; the gap must stay above zero
RenewingThe grant covers records created after signing, delivered on a scheduleThe grant covers a fixed archive onlyCount renewals and refresh deliveries by dated cohort, not renewals promised
ScarceFew other holders could grant comparable rights for the same buyer taskThe buyer can license similar records from many holdersList every other holder a buyer could approach; more than a handful is a warning

Enforceability: what courts and statutes let a rights holder stop

Enforceability depends on where the material lives. Rights over published content are only as strong as the law on mining and fair use in each jurisdiction; rights over private business records rest mainly on contracts and access control, because rivals cannot lawfully reach the records in the first place.

The clearest recent case shows enforcement working for a curated collection. In Thomson Reuters v. Ross Intelligence, the District of Delaware's February 2025 opinion granted Thomson Reuters partial summary judgment on direct infringement of Westlaw headnotes and rejected Ross's fair use defense. According to the opinion, Ross obtained about 25,000 Bulk Memos from LegalEase after Thomson Reuters declined to license its content to a competitor. It is a trial-court ruling, and it can still change on appeal.

The European Union draws the line differently for public content. Article 4 of Directive (EU) 2019/790 allows text and data mining of lawfully accessible works unless the rightholder has expressly reserved that use in an appropriate manner, such as machine-readable means for content made publicly available online. A publisher's right is defensible there only if it is reserved properly, while a company's private records are not lawfully accessible to outsiders at all.

In the United States the policy argument is still open. The Copyright Office's May 2025 pre-publication report on generative AI training records commenters split between those who argued that licensing requirements would stifle development and those who argued that unlicensed training usurps an emerging market for works as training material. Score enforceability per jurisdiction, and never higher than the weakest market the buyer trains in.

Necessity and scarcity: could the buyer get the same result elsewhere?

Often, yes, and that is where most rights-based moat claims break. A buyer that can pass its acceptance test with synthetic, public or in-house data does not need the right, and a buyer that can license comparable records from many holders does not need any one holder.

Model developers now produce much of their own training data. Meta's Llama 3.1 model card describes fine-tuning data that includes over 25 million synthetically generated examples, and its Llama 4 card describes pretraining on a mix of publicly available data, licensed data and information from Meta's own products and services. Each of those is a substitute a licensor has to beat on the buyer's own test, not on a slide.

Scarcity fails quietly when a category has many holders. News Corp is not the only publisher with an OpenAI partnership: OpenAI described its Axios partnership as expanding its work with the news industry. When a buyer can assemble a category from many holders, each holder's rights earn a price but rarely a moat. Business records can be more fragmented still: every regional services firm keeps dispatch records, so scarcity comes from depth, linked outcomes and rare cases rather than from the record type.

Exclusivity changes scarcity only for its term and only for the licensor who granted it. It does nothing about a comparable flow held by someone else, which is why the scarce link asks who else could sell, not what the contract forbids.

The scorecard separates rights positions that look alike in a pitch. In the three illustrative cases below, only the one with no link at 0 supports a moat claim, and the case with exclusivity fails.

Illustrative case A: a data company holds a three-year grant to new dispatch and outcome records from a field services firm, signed by an officer with documented authority, for training and evaluation. Illustrative case B: a company licenses a fixed archive of published articles that are freely readable online and carry no machine-readable reservation. Illustrative case C: a company holds an exclusive license to a decade of claims notes, but the grantor held them as a service provider for its insurer clients and had no authority to license them.

Case C is the warning. Exclusivity raised its scarcity score, but missing authority sets the verdict, and a buyer that trained on those records would inherit the defect. Case B is not worthless; it is a one-time license priced as one. The figure draws each case as a chain, with the broken links marked.

Three illustrative rights positions, scored link by link
LinkCase A: flow grant, private recordsCase B: archive of public articlesCase C: exclusive, no authority
Valid2: signed grant, documented authority2: the publisher owns the archive0: grantor held the records for clients
Enforceable2: records are private; the contract limits use1: open to mining where no reservation exists1: exclusivity binds only the grantor
Necessary1: beats substitutes on rare outcome cases1: overlaps heavily with public text2: buyers need claims reasoning
Renewing2: new records delivered each quarter0: fixed archive, no new records1: renewal possible once authority is fixed
Scarce1: a few comparable firms exist0: many publishers offer similar text2: few holders of this depth
Verdict (lowest link)1: defensible, watch scarcity0: price it as a one-time license0: a liability until authority is fixed
Three illustrative rights positions drawn as five-link chains, valid, enforceable, necessary, renewing and scarce, with case A unbroken, case B broken at renewing and scarce, and case C broken at valid.
The five-link rights scorecard applied to three cases. Illustrative: scores are examples, not measurements of real companies.

How to collect evidence that could disprove the claim

Build a small, permissioned evidence set around the claim before scoring it, and record the cases you excluded and the one that most challenges the conclusion. SourceX has not run or published this study; the design is here so the same test can be applied to any company, SourceX included.

At a glance

How to collect evidence that could disprove the claim

Results are not yet published. A company that cannot assemble these records for its own grants has answered the defensibility question in a different way.

Checklist · 7 items

  1. Sampling frame: every active rights grant the company holds, listed before any are selected, with grant date, term, scope and counterparty type.
  2. Sample: a seeded random draw of grants stratified by size and age, plus every grant the company describes as exclusive.
  3. Replacement cost: for each sampled grant, the steps and elapsed time a funded rival would need to obtain comparable rights from another holder.
  4. Legal scope: counsel's reading of what each grant permits and excludes, and which jurisdictions' mining or fair use rules could make it unnecessary.
  5. Technical utility: the buyer's acceptance test run with and without the licensed data, against the best available substitute.
  6. Repeat behavior: renewals, refresh deliveries and expansions by buyer and dated cohort, with lapsed grants kept in the denominator.
  7. Challenge case: the sampled grant with the lowest verdict, reported next to the headline score with the reason it failed.

0 of 7 ticked. Tick each item as you go.

What could make a rights-based moat thesis wrong

Four forces can erode a rights position after it has been scored. Name the one a company is most exposed to before valuing its moat, and rescore on a schedule rather than once at the term sheet.

  • Law moves toward free use. A broader reading of fair use in the United States, or of the mining exception in the European Union, would weaken rights over published content; Ross is a single trial-court ruling, and the Copyright Office report shows how divided the arguments remain.
  • Substitutes improve. Every gain in synthetic data, such as the over 25 million synthetic examples in Meta's Llama 3.1 fine-tuning data, lowers the necessity score of rights to routine material.
  • Buyers stay few. Reddit's annual report for 2024 ties substantially all of its licensing contract value to two partners, so a renewing right sold to two buyers is a concentration risk as much as a moat.
  • Buyers supply themselves. Llama 4's pretraining drew on information from Meta's own products and services, and a buyer with its own flow can stop renewing an outside one.

Questions to ask about any rights-based moat claim, SourceX included

Ask for the weakest link first, with evidence, and score it with your own definitions. Hold SourceX to the same standard.

SourceX works with the companies that hold business records, turning selected records into licensed AI datasets; its part covers supplier sourcing and qualification, a rights review, the contract, data preparation, delivery and payment. In rights terms the agreement comes first: the supplying company approves scope and terms and signs before a single record is delivered, and each deal's signed agreement states its exclusivity and term instead of leaving them to a general promise.

Deal and commercial metrics are not published; SourceX makes company and commercial metrics available to investors directly. It is raising capital under Rule 506(c) from accredited investors, and the investors page takes requests for materials.

Checklist · 5 items

  1. Which of the five links is weakest across your active grants, and what evidence did you use to score it?
  2. What share of your grants cover records created after signing, and how many have renewed on the record?
  3. For your largest grant, who else could license comparable records to the same buyer?
  4. Which jurisdiction's mining or fair use rule would most reduce the need for your rights?
  5. Which grant would fail the validity test today, and what would it take to fix it?

0 of 5 ticked. Tick each item as you go.

Methods note: how the scorecard was built and what it does not measure

The five-link scorecard, the lowest-link rule, the illustrative cases and the evidence design are this article's own editorial frameworks, assembled on 2026-10-26 from the District of Delaware's February 2025 opinion in Thomson Reuters v. Ross Intelligence, Article 4 of Directive (EU) 2019/790, the Copyright Office's May 2025 pre-publication report, OpenAI's News Corp and Axios partnership announcements, Reddit's S-1 and annual report for 2024, and Meta's Llama 3.1 and Llama 4 model cards, read directly on GitHub, each checked that day.

Not measured: the link scores of any real company, the value of any rights grant, renewal rates across the market, and SourceX's own grants or deals. The three cases are illustrative and describe no real company or transaction.

Frequently asked questions

What is the difference between data rights and a data moat?

Data rights are legal permissions to use records for stated purposes. A data moat is an economic position in which buyers keep coming back and rivals cannot easily meet their need another way. Rights can support a moat, but only when they are valid, enforceable, necessary, renewing and scarce at the same time.

Does an exclusive data license create a moat?

Not by itself. Exclusivity stops the licensor from granting the same rights to others for the term, but it does not stop a buyer from using synthetic or public substitutes or licensing comparable records from another holder. It also cannot repair a grant made by someone who lacked authority to make it.

Can fair use or text and data mining exceptions make data rights worthless?

They can reduce the value of rights over published content in the jurisdictions where they apply. Article 4 of the EU copyright directive allows mining of lawfully accessible works unless the rightholder reserves the use, while the Ross ruling in Delaware rejected fair use for copied headnotes. Private business records are different because outsiders cannot lawfully reach them, so their protection rests on contracts and access control. This is general information, not legal advice.

Why do flow rights matter more than archive rights to investors?

A buyer that has trained on an archive keeps what its model learned, so the archive is usually sold once. A right to records created after signing gives the buyer something new each period, which is what makes renewals, and therefore repeat demand, possible. Renewals on the record are the evidence that a flow right is working.

Does SourceX publish its rights or deal metrics?

No. Rights and deal metrics are not public, and SourceX gives company and commercial metrics to investors directly. Exclusivity and term are fixed deal by deal in each signed agreement. Investor materials can be requested at sourcex.si/investors, and any offering by SourceX goes only to verified accredited investors, through its offering documents.

Sources

  • In Thomson Reuters v. Ross Intelligence (D. Del., February 2025), the court granted partial summary judgment on direct infringement of Westlaw headnotes, rejected fair use, and recounted that Ross obtained about 25,000 Bulk Memos from LegalEase after Thomson Reuters declined to license. Source
  • Article 4 of Directive (EU) 2019/790 provides a text and data mining exception for lawfully accessible works that applies unless rightholders have expressly reserved the use in an appropriate manner, such as machine-readable means for content made publicly available online. Source
  • The U.S. Copyright Office's May 2025 pre-publication report on generative AI training records commenters split between arguments that licensing requirements would stifle development and that unlicensed training usurps an emerging market for works as training material. Source
  • OpenAI's multi-year global partnership with News Corp gives OpenAI access to current and archived content from News Corp's major news and information publications. Source
  • OpenAI described its partnership with Axios as expanding OpenAI's work with the news industry. Source
  • Reddit's S-1 disclosed data licensing arrangements entered in January 2024 with terms ranging from two to three years. Source
  • Reddit's annual report for 2024 says substantially all of the contract value associated with its licensing revenue comes from two partners and that the arrangements may not be renewed or may be renewed on less favorable terms. Source
  • Meta's Llama 3.1 model card reports fine-tuning data that includes over 25 million synthetically generated examples. Source
  • Meta's Llama 4 model card describes pretraining on a mix of publicly available data, licensed data and information from Meta's products and services. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies