Skip to content

Leadership and readiness

Data supplier permission scope: a four-axis map for investor diligence

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

Data supplier permission scope is what a supplier has actually authorized across four axes: which datasets, for which uses, by which recipients, over which period. Supply is sellable only where all four overlap a buyer's need. Investors should ask a data company for a scope map that places every supplier grant and buyer license on those axes, then sample deliveries to confirm each sits inside its narrowest upstream grant.

Checklist

Data supplier permission

This page covers data supplier permission scope: a four-axis map for investor diligence.Illustrative

Key takeaways

  • Permission has four dimensions, dataset, use, recipient and time, and a grant that is broad on three can still block a sale on the fourth.
  • Scope only narrows as data moves downstream, so a data company can sell no more than the narrowest grant behind each delivery.
  • Licensing provenance decays in practice: the Data Provenance Initiative found licenses unspecified for more than 70 percent of popular datasets it traced.
  • A scope-fit rate on sampled deliveries, with unknowns counted as unknown, tests whether counted supply was sellable.
  • Stated buyer plans, such as micro1's announced $1 billion data program, raise the stakes on the use axis but are plans, not purchases.

What is data supplier permission scope, and why does it have four dimensions?

Data supplier permission scope is the set of datasets, uses, recipients and time periods that a company supplying records has actually authorized. It has four dimensions because a buyer's need has four: which records, what the buyer will do with them, who will handle them and for how long. A sale works only where the supplier's grant covers all four at once.

That makes scope an asset-quality question for an investor. A data company can report many signed supplier agreements and still be unable to fill a buyer's request, because each agreement may cover the wrong use, exclude the buyer's contractors or lapse before the buyer's model ships. Counting signatures measures willingness to license; mapping scope measures what can be sold.

Demand makes the map urgent. In October 2026 micro1 announced a plan to put $1 billion over the next year into buying and licensing companies' operational data, which it turns into reinforcement learning environments that mimic how real businesses run. That is a stated plan, not money spent, and it does not show that any given company's records will be bought. It does name a use, simulated environments for reinforcement learning, that many older data grants were never drafted to cover.

The four-axis scope map, axis by axis

Draw one map per dataset, with four axes and two marks on each: what the supplier granted and what the buyer requests. Sellable scope is the overlap on every axis; anything the buyer needs outside the grant is a gap that blocks the sale or calls for a new signature.

At a glance

The four-axis scope map, axis by axis

Scope also narrows along the chain. Individuals and customers set the outer limit through notices, consents and contracts; the supplying company grants within that limit; the data company sublicenses within the supplier's grant; and the buyer's affiliates and contractors sit inside the buyer's license. No link can pass on more than it received, so the narrowest grant behind a delivery sets what the whole chain can sell.

Illustrative: a supplier grant covers support tickets, notes and attachments, for evaluation, training and fine-tuning, by a named licensee and its affiliates, for records created before signing and use during the term. A buyer asks for tickets and notes, for training, fine-tuning and reinforcement learning environments, by the buyer and its contractors, with use continuing after the term ends. The map shows three gaps, one each on use, recipient and time, and none on the dataset axis. The four axes are:

  • Dataset: which systems, record types, fields and creation periods are covered, and what was carved out. Evidence: the scope schedule and the manifest that implements it.
  • Use: what the recipient may do, such as evaluate, train, fine-tune, build reinforcement learning environments or distill into another model. Evidence: field-of-use wording, read against the buyer's stated purpose.
  • Recipient: who may receive or handle the data, from the named licensee to affiliates, contractors and sublicensees. Evidence: sublicensing and affiliate clauses beside the buyer contract's list of permitted parties.
  • Time: which record periods are covered, how long use may continue, what survives termination, and whether earlier notices allowed the use at all. Evidence: term and survival clauses, and the notice versions in force when the records were created.
A four-row scope map for dataset, use, recipient and time, with cells marked sellable, granted only, or requested but not granted; the gaps fall on reinforcement learning environments, contractors and use after the term.
An illustrative supplier grant against a buyer request: the narrowest axis sets what can be sold.

Where permission narrows between the source and the buyer

It narrows wherever a promise made upstream limits a use downstream, and the losses are easy to miss because they rarely change the record count. Four public sources show the main mechanisms.

Purpose limits bind personal data. Under Article 5(1)(b) of the GDPR, personal data are to be "collected for specified, explicit and legitimate purposes and not further processed in a manner that is incompatible with those purposes". Where EU personal data sit inside business records, the use axis starts from the purposes stated at collection, whatever a later supplier agreement says.

Retroactive edits do not reliably widen scope. Staff at the U.S. Federal Trade Commission warned in February 2024 that a company risks acting unfairly or deceptively if it loosens its data practices, for instance by starting to share data with third parties or to train AI on it, and reveals the shift only through a quiet, after-the-fact edit to its terms or privacy policy. A supplier whose notice changed last year cannot assume the new wording reaches records collected under the old one, which makes notice history a time-axis question.

Licensing provenance decays as data moves, and permission changes over time. When the Data Provenance Initiative traced licenses across widely used text datasets, it found licenses left unspecified for more than 70 percent of popular datasets on GitHub and Hugging Face, with labeling error rates above half. Across 14,000 web domains, the Consent in Crisis researchers saw restrictions added during 2023-2024 close off more than 5 percent of all tokens in the C4 corpus, and more than 28 percent of its most actively maintained critical sources, entirely. Private records are not web pages, but both studies show that permission recorded once and never re-checked goes stale.

Measuring scope fit: a reconciliation an investor can reproduce

Test scope on deliveries, not on agreements: sample delivered datasets or license line items, trace every source record set back to its grant, and score each delivery on all four axes. The headline metric is a scope-fit rate, with misfits and unknowns reported by axis.

Denominators decide whether the numbers mean anything. Freeze the list of deliveries in a stated period before sampling, draw at random within buyer and dataset strata, and count a delivery whose grant cannot be found as unknown, not as a fit. A company that cannot produce the grant behind a delivery has a records problem whatever its contracts say.

These are the metrics to ask for. SourceX has not published values for them and none are reported here; each is defined with its denominator:

Measuring scope fit: a reconciliation an investor can reproduce
MetricDefinitionDenominatorWhat it reveals
Scope-fit rateDeliveries inside the narrowest upstream grant on all four axesSampled deliveries from a frozen periodWhether counted supply was sellable as delivered
Unknown-scope shareDeliveries whose upstream grant could not be producedThe same sampleWhether the company can prove scope at all
Misfit by axisMisfits split by dataset, use, recipient and timeMisfit deliveriesWhich axis binds, and so which fix matters
Use headroomSupply whose grant covers a named use, such as reinforcement learning environmentsAll supply the company countsHow much supply can meet a new kind of demand
Time-to-expiry profileSupply grouped by remaining term and survival rightsAll supply the company countsHow fast sellable scope shrinks without renewals

The dataset diligence evidence request list for permission scope

Ask for documents that let you rebuild the scope map yourself, then for a sample that tests it. Executed documents matter more than templates, because amendments, side letters and carve-outs live in the executed set.

Tie every request to a named axis; a request that maps to no axis tends to produce paper that proves nothing about scope. The list runs from the source outward:

The dataset diligence evidence request list for permission scope
RequestAxis testedWhat good looks likeRed flag
Notice and consent versions in force when each dataset's records were createdTime and useVersions dated and mapped to record periodsOne current notice offered for every period
Executed supplier grants with scope schedulesDataset and useSystems, fields, periods and carve-outs listedTemplates only, or schedules that say all data
Sublicensing and affiliate clauses beside each buyer's permitted partiesRecipientEvery buyer party traced to a clauseContractors or affiliates missing from the grant
Buyer license line items with field-of-use wordingUseUses no wider than the upstream grantBuyer uses broader than any grant
Delivery manifests linked to grant IDsDatasetEach delivered file traced to one grantFiles with no grant reference
Term, survival and withdrawal logTimeExpiries and withdrawals dated and applied to deliveriesNo log, or withdrawals that never reach delivered data
Remedy terms for out-of-scope use, including derived modelsAll axesRemedies stated and weighed in the risk reviewSilence on models already trained
A stratified sample of deliveries re-traced by a reviewer the investor picksAll axesScope-fit and unknown shares with denominatorsA sample chosen by the company

Can one signed grant cover a buyer's whole request?

Rarely, which is why a signed agreement or a polished sample is the strongest alternative explanation for apparent supply quality. A signed grant shows that a supplier agreed to something; it does not show that the something matches what a paying buyer asks for. A clean sample shows the records look good, not that they may be used the way the buyer intends.

The best case for counting signatures is that scope can be widened later by amendment, so signed suppliers are an option on future scope. That holds when the supplier is still reachable, the upstream notices allow the wider use, and the buyer will wait. Each condition is a fact to check, which is why the scope map records gaps as gaps rather than as pending amendments.

The test is direct: pick three recent buyer requests and ask the company to show, for each, the grants that covered every axis at the time of delivery. If it can, its signatures are worth counting. If it cannot, the count describes relationships, not sellable supply.

Where the scope-map thesis could be wrong

The thesis is that sellable scope, not signed supply, drives value in a data company. Name the ways it could fail before relying on it; each comes with a source or a test an investor can track.

  • Remedies can reach the model, not just the dataset. The FTC's 2021 final order against Everalbum required the company to delete models and algorithms it developed using photos and videos uploaded by its users. A scope error found after training can cost far more than the data did.
  • Stated demand may not arrive. micro1's $1 billion program is a published plan; it does not show that any particular company's records will be bought, and use headroom for reinforcement learning environments is worth only what buyers actually pay for.
  • Maps go stale. Notices change, grants expire and withdrawals arrive, as the 2023-2024 rise in web restrictions shows, so a scope map without a dated refresh is a snapshot rather than an asset register.
  • The law differs by record type and place. Purpose limits under the GDPR apply to personal data; records with no personal data, or held outside its reach, face different rules, so one company's map can be much broader or narrower than another's for reasons unrelated to execution.
  • Narrow scope can be the right product. Some buyers want one use by one recipient, and a company whose tightly drawn grants match those buyers may do better than one with broad grants and no buyers.

What to ask a data company about its scope map, SourceX included

Ask the same questions of every company that sells licensed data, using definitions you set. SourceX is one application of the framework. It handles licensing deals end to end for companies that license operational records to AI developers, from finding and qualifying suppliers and reviewing rights to the contract, data preparation, delivery and payment. The supplying company agrees the scope and terms, and signs, before any records reach SourceX; from then on, SourceX may use, sublicense and distribute the de-identified dataset within what that agreement permits.

SourceX keeps company and commercial figures off its public pages and gives them to investors directly, so none appear here. It is raising capital from accredited investors under Rule 506(c); materials go out on request through its investor page, and only verified accredited investors can take part in any offering, which is made through its offering documents. The questions below can be put to its team as to any other:

Checklist · 6 items

  1. Can you show the scope map for your three largest datasets, with the grant behind each mark?
  2. What was your scope-fit rate on deliveries over the last four quarters, and how many deliveries were unknown?
  3. Which axis causes most misfits, and what changed because of it?
  4. Which uses, such as reinforcement learning environments or distillation, do your grants cover, and for what share of the supply you count?
  5. How do you map notice versions to record periods, and who checks the mapping?
  6. What happens to delivered data and derived models when a supplier withdraws or a grant expires?

0 of 6 ticked. Tick each item as you go.

Methods note: sources, date and what the scope map leaves unmeasured

The scope map, its five metrics and the evidence request list were drawn up for this article as editorial frameworks, dated 2026-10-27. Their evidence comes from the GDPR's purpose limitation principle, an FTC staff post (February 2024) and the FTC's Everalbum order (2021), the Data Provenance Initiative and Consent in Crisis audits, and micro1's October 2026 announcement. The supplier-and-buyer example is illustrative and describes no real company or deal.

Not measured: scope-fit rates, unknown-scope shares or use headroom at SourceX or any other company, the share of real grants that cover any given use, and the effect of scope breadth on price. Each needs company records that have not been published, and none of the frameworks has yet been tested against a sample of executed agreements.

Frequently asked questions

What does permission scope mean for a data supplier?

It is the combination of datasets, uses, recipients and time periods a supplier has authorized. A grant can be broad on some of these and narrow on others, and a buyer's request can be filled only where the grant covers every dimension the buyer needs.

Why is a signed supplier agreement not enough for investors?

A signature shows that a supplier agreed to license something, not that the grant matches what buyers want. The use may be too narrow, the buyer's contractors may be excluded, or the term may end before the buyer's model ships. Investors should test grants against actual deliveries.

How can an investor test a data company's permission scope?

Ask for a scope map per dataset and a frozen list of recent deliveries, then have a reviewer you choose trace a stratified sample back to the grants behind them. Report a scope-fit rate, the share of deliveries whose grant could not be found, and misfits by axis.

Can a company widen permission by updating its privacy notice?

Not reliably for data it already holds. FTC staff cautioned in 2024 that loosening data practices, for example to train AI, and announcing it only by quietly rewriting terms or a privacy policy after the fact can be unfair or deceptive. Counsel should review the notice history for the specific records.

Does SourceX publish scope-fit or supply metrics?

No. SourceX does not publish commercial figures and gives them to investors directly. Investors can ask for materials through sourcex.si/investors and put scope questions to the team there; participation in any offering is limited to verified accredited investors.

Sources

  • micro1 says it plans to spend $1 billion over 12 months acquiring and licensing enterprise operational data to build reinforcement learning environments (October 2026; a stated plan). Source
  • GDPR Article 5(1)(b): personal data collected for specified, explicit and legitimate purposes and not further processed incompatibly with them. Source
  • FTC staff (February 2024): adopting more permissive data practices, such as AI training, disclosed only by a surreptitious, retroactive terms or policy change may be unfair or deceptive. Source
  • The FTC's May 2021 final order required Everalbum to delete models and algorithms developed using users' photos and videos. Source
  • The Data Provenance Initiative audit found licenses unspecified for 70 percent or more of popular datasets on GitHub and Hugging Face, with error rates of 50 percent or more. Source
  • Consent in Crisis audited 14,000 web domains; over 2023-2024, about 5 percent or more of C4 tokens and 28 percent or more of its critical sources became fully restricted. 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