Deal economics
Repeat supplier qualification economics: price evidence by shelf life
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Repeat supplier qualification economics turn on evidence shelf life. A returning supplier’s entity, signatories and master agreement can be reused until something material changes; the new dataset’s rights, its overlap with earlier exclusive licenses and its buyer fit cannot. Web certificate rules split evidence the same way. In this article’s illustrative model, a repeat deal still costs 56.7 percent of a first, because most of its hours are fresh checks.
Value drivers
Repeat supplier
Key takeaways
- A repeat data deal is a different bundle of checks, not a discount: company-level evidence reuses, dataset-level evidence starts from zero.
- Certificate rules allow identity checks to be reused for 398 days and domain control for 200 days now, falling to 10 days in 2029.
- In the illustrative ledger, a repeat deal costs 34 of a first deal’s 60 hours, and 27 of those 34 are fresh dataset checks.
- A change of control pushes a repeat deal to 80.0 percent of a first; the ceiling on capacity from returning suppliers is about 1.76 times.
- Returning suppliers are self-selected, so only matched, class-by-class comparisons show whether a saving comes from reuse.
What drives repeat supplier qualification economics?
The share of qualification evidence that is still true when the supplier comes back. A returning supplier does not get a percentage off its first review; it gets a different bundle of checks. Some evidence describes the company and changes rarely, so it can be reused after a short refresh. Some describes the new dataset and has never been checked before, so it costs as much as it did the first time, or more.
That split decides whether returning suppliers give a data licensing company operating leverage. If most qualification hours sit in company-level evidence, every second dataset is cheap and capacity compounds. If most sit in dataset-level checks, a repeat deal saves less than a sales deck implies, and the saving can vanish when a supplier changes owners, systems or privacy terms.
This article prices that split. It borrows the clearest public example of evidence reuse rules, from the web certificate industry, builds an illustrative evidence-class model of a first and a repeat data deal, tests the obvious objection, and sets out what an investor should ask to see.
Evidence has a shelf life: what certificate authorities may reuse, and for how long
Mature verification regimes put an expiry date on each kind of checked evidence, and the dates differ by how fast the fact can change. The CA/Browser Forum’s Baseline Requirements, which govern publicly trusted web certificates, let a certificate authority reuse earlier validations only within set windows and say plainly that “In no case may a prior validation be reused if any data or document used in the prior validation was obtained more than the maximum time permitted for reuse of the data or document prior to issuing the Certificate.”
The windows have been cut twice, and the cuts fell hardest on the fact that changes fastest. Version 1.7.4 of April 2021 allowed one 825-day window for all validations. Ballot SC042 (version 1.7.5) cut domain and IP address validation to 398 days from 1 October 2021. Ballot SC081, adopted on 11 April 2025, kept identity separate and put domain control on a schedule: 398 days for identity and 200 days for domain control from 15 March 2026, 100 days for domain control from March 2027 and 10 days from March 2029. Some checks are never reused at all: a CAA record must be processed within its time-to-live or 8 hours of issuance, whichever is greater.
Reuse also has to be provable. From 15 July 2026, the requirements say verification records must include at a minimum the information validated and the validation method used. Requalification of the authorities themselves is continuous rather than episodic: Mozilla’s root store policy requires audits before inclusion and at least annually after, with successive period-of-time audits contiguous and without gaps. Management-system certification sets a similar ratio: IAF MD 5 puts annual surveillance at about one-third of initial audit time and recertification at about two-thirds of what an initial audit would need at that point. The certificate rules, read from the document itself:
| Evidence in the certificate rules | Maximum reuse | From | Where it is set |
|---|---|---|---|
| Subject identity information (organization details) | 825 days, then 398 days | 398 days for certificates issued from 15 March 2026 | Section 4.2.1 table |
| Domain name and IP address validation | 398, then 200, 100 and 10 days | 200 days from 15 March 2026; 100 from 15 March 2027; 10 from 15 March 2029 | Section 4.2.1 table |
| CAA record check | Not reused: within the record’s time-to-live or 8 hours, whichever is greater | Current text | Section 4.2.2.1 |
| Cached WHOIS or RDAP bootstrap data | At most 48 hours old | Current text | Section 3.2.2.4.12 |
| Record of what was verified and how | Kept for every verification | Effective 15 July 2026 | Section 5.4.1 |
Which supplier evidence carries over to a second dataset, and which never does
For a data supplier, company-level evidence carries over and dataset-level evidence does not. The entity, its control and its signatories can be reused after a refresh check; a signed master agreement can be reused until something material changes; the inventory of systems and the company’s privacy and contract policies need a check that they still hold. The rights in the newly selected records, the scope of the new dataset and the buyer’s fit test start from zero.
One check gets harder on the second deal. A supplier that granted an exclusive license for its first selection cannot offer the same records again for AI training, so the second scope has to be checked for overlap against what was already selected and delivered. In this model that check takes 9 hours on a repeat deal against 6 on the first.
The ledger below is the article’s editable model. Hours are illustrative inputs, not measurements, and the reuse rules follow the logic of the certificate table: reuse what describes the company, refresh what can drift, and check fresh what is new. The ledger:
| Evidence class | Shelf life on a repeat deal | First deal, hours | Repeat deal, hours | Redone in full when |
|---|---|---|---|---|
| Entity, control and signing authority | Refresh check | 6 | 1 | Control or a signatory changed |
| Master agreement | Reused until a material change | 10 | 0 | Control changed (an amendment, 4 hours) |
| Systems inventory and export route | Refresh check | 12 | 3 | A new source system is involved |
| Company privacy and contract policies | Refresh check | 8 | 3 | A notice or contract template changed |
| Rights in the selected records | Fresh every deal | 10 | 10 | Always |
| Scope and overlap with earlier licenses | Fresh every deal | 6 | 9 | Always |
| Sample and buyer-fit test | Fresh every deal | 8 | 8 | Always |
| Total (illustrative) | 7 of 34 repeat hours are reuse or refresh | 60 | 34 | See the scenarios below |
The requalification model: what a returning supplier saves, and when the saving shrinks
In this model a returning supplier costs 34 of the first deal’s 60 hours, 56.7 percent, because 27 of the 34 are checks that must be fresh on every dataset. The saving is large at the company level, where 36 first-deal hours shrink to 7 (the master agreement from 10 hours to none, the three refreshed classes from 26 to 7), and slightly negative at the dataset level, where 24 hours become 27.
The saving shrinks as soon as something material changes. A changed privacy notice or customer contract template pushes the repeat deal to 39 hours (65.0 percent of the first), a new source system to 43 hours (71.7 percent) and a change of control to 48 hours (80.0 percent), because the entity check is redone and the master agreement needs an amendment. For comparison, management-system certification budgets about one-third of initial audit time for a surveillance audit and about two-thirds for a recertification.
For an investor, the test of a returning-supplier claim is therefore not the discount itself but the mix behind it: how many repeat deals arrive with nothing changed, and how many carry a change that resets part of the ledger. The scenarios:
| Deal (illustrative) | Qualification hours | Share of a first deal | Datasets per 600 team hours |
|---|---|---|---|
| First deal | 60 | 100.0% | 10.0 |
| Repeat, nothing changed | 34 | 56.7% | 17.6 |
| Repeat, privacy or contract policy changed | 39 | 65.0% | 15.4 |
| Repeat, new source system | 43 | 71.7% | 14.0 |
| Repeat, control changed | 48 | 80.0% | 12.5 |
How much capacity can returning suppliers add?
Returning suppliers raise qualification capacity, but by a bounded amount. Assume a team with 600 practical qualification hours a quarter and a repeat-deal mix in which 60 percent arrive with nothing changed, 20 percent with a policy change, 10 percent with a new source system and 10 percent with a change of control. A repeat deal then averages 37.3 hours.
Under that mix, moving from no returning suppliers to all returning suppliers lifts capacity from 10.0 to 16.1 datasets a quarter, 1.61 times. Even if nothing ever changed, the ceiling would be 60 divided by 34, or 1.76 times, because the fresh dataset checks never shrink. A company claiming that repeat supply multiplies throughput well beyond that should be able to show which evidence it reuses that this model does not. Capacity by returning share:
- No returning suppliers: 60.0 hours per dataset, 10.0 datasets per 600 hours.
- One in four datasets from returning suppliers: 54.3 hours, 11.0 datasets.
- Half from returning suppliers: 48.7 hours, 12.3 datasets.
- Three in four from returning suppliers: 43.0 hours, 14.0 datasets.
- All from returning suppliers: 37.3 hours, 16.1 datasets.
Is the saving real, or did only the easy suppliers come back?
Possibly not, and the model cannot tell. Returning suppliers are not a random draw: the ones that come back are often those whose first deal went smoothly, with clean rights, one system and an engaged sponsor. A company comparing its average first-deal hours with its average repeat hours will see a saving even if reuse saved nothing, because the hard suppliers never returned.
The fair comparison is matched: the same supplier’s first and second deals, class by class, with suppliers who have not yet returned kept in the cohort as not yet observed rather than dropped. If the saving sits in the reuse and refresh rows, it is reuse. If it sits in the fresh rows, the second dataset was simply easier.
The opposite error is cheap requalification. Skipping the three refresh checks cuts a repeat deal from 34 to 27 hours, but a deal built on stale company-level evidence has to be unwound. With an illustrative 40 hours to unwind, skipping pays only if fewer than 17.5 percent of returning suppliers have an undetected material change, and the expected hours at a 15 percent rate, 33.0, barely beat the refresh while adding rights risk the hours do not price. A related trap: more repeat deals add expense without adding licensable supply when they relist records that overlap an earlier exclusive selection.
Measuring requalification in a supplier cohort
Track requalification per evidence class in a dated supplier cohort, with the reused evidence and its original date recorded, as the certificate rules now require for verification records. The design below is a method, and SourceX has not published results for it. The fields and rules:
At a glance
Measuring requalification in a supplier cohort
- Cohort: suppliers grouped by the quarter their first dataset was accepted; each later dataset is linked to the supplier and to its first deal.
- Per class, per deal: hours, whether the evidence was reused, refreshed or checked fresh, the date the reused evidence was first verified, and who verified it.
- Refresh outcomes: the share of refresh checks that found a change, by class, and what the change reset.
- Stale-evidence failures: problems found after a deal that trace to reused evidence, such as an undisclosed change of control or an overlapping selection, with their unwinding hours.
- Denominators: every first-deal supplier in the cohort, with those yet to return counted as not yet observed, never dropped.
- Comparison: same-supplier first and second deals, class by class, reported beside the unmatched averages.
What could make the requalification thesis wrong
Five developments could shrink the value of returning suppliers. Each would show up in the cohort fields above before it shows up in revenue.
- Archives run out. If the first license is exclusive for AI training over the selected records, a second deal needs records that were not selected or that are new, and many companies hold only one archive worth licensing.
- Reuse windows shorten. The certificate industry cut its domain validation window from 825 days to a scheduled 10 over eight years; buyers or regulators could likewise demand fresher rights and privacy evidence, moving hours from the reuse rows to the fresh rows.
- Control changes reset the ledger. Acquisitions and restructurings send entity, agreement and policy evidence back to first-deal cost, and in this model a change of control costs 80.0 percent of a first deal.
- Buyer fit does not repeat. The fit test is fresh every time; if buyers want different records from the same supplier, the second deal’s scope work grows.
- Skipped refresh checks. A company that shows cheap repeat deals by skipping refresh checks carries rights risk that only appears later, in unwinding hours or disputes.
Requalification questions for any data company, SourceX included
Ask any data licensing company to show its requalification ledger by evidence class, not its average repeat-deal cost. SourceX should face the same questions, and its published terms make a returning supplier’s path easy to test.
SourceX’s how-it-works page says a supplier signs one Master Supplier Licensing Agreement after qualification that also covers later deals, that returning suppliers reuse the Master and sign a new Dataset Deal Confirmation, and that a material change needs a signed amendment. Its investor page says one repeatable agreement shortens legal review for both sides on every deal. Under the same terms, exclusivity covers only the selected data, not future data or unselected records, and the supplier confirms the data has not been licensed for AI training before, which is why the overlap check is fresh on every deal.
SourceX has published no requalification figures, and none are stated here; according to its investor page, company and commercial metrics are shared only with investors. Its raise is limited to accredited investors under Rule 506(c), and the investors page takes requests for materials. Questions to bring:
- What share of your qualification hours sits in company-level evidence, and what share in dataset-level checks?
- When a returning supplier signs a new confirmation, which checks do you reuse, which do you refresh, and what dates do you record for each?
- How often does a refresh check find a change of control, signatory, system or policy, and what does each reset?
- How do you check a second selection for overlap with records already licensed exclusively?
- Are your repeat-deal savings measured on matched suppliers, with suppliers who have not returned kept in the denominator?
- Which stale-evidence failures have you found after a deal, and how many hours did each take to unwind?
Methods note: the reuse table, the model and what was not measured
Reuse table: read on 28 October 2026 from docs/BR.md in a clone of github.com/cabforum/servercert at commit 3574ae8 (4 October 2026, version 2.3.1), with earlier wording from the tags BRs/v1.7.4 (commit 6e98f8a) and BRs/v1.7.5 (commit 8333235). A short script, br_reuse_history.py, runs git show for every tag named BRs/* and for HEAD, extracts section 4.2.1 from its heading to the next one, and lists every figure written as a number followed by days; the CAA, WHOIS and logging rules were read from the current text. The tagged versions start at 1.7.4, so changes before April 2021 are not covered.
Model: an editorial model built for this article on 28 October 2026 and computed by xs0184_requalification_model.py with exact fractions, rounding half up once at print. Every input is illustrative: the hours in the ledger; the scenario changes (policy change sets policies to 8 hours, a new source system sets systems to 12, a change of control sets entity to 6, the agreement to a 4-hour amendment and policies to 8); 600 practical team hours a quarter; the repeat mix of 60, 20, 10 and 10 percent; and 40 unwinding hours with a 15 percent change rate for the skipped-refresh case. Classes are treated as independent and additive.
Other evidence: Mozilla’s root store policy at commit bb613bb (read from a clone) and IAF MD 5, issues of 2019 and 2023, checked through search excerpts on iaf.nu; SourceX’s how-it-works and investor pages were fetched directly. Not measured: real qualification hours, reuse rates or refresh outcomes at SourceX or any data company, how often suppliers return, and how much of any observed saving reflects selection rather than reuse.
Frequently asked questions
What is supplier requalification in AI data licensing?
It is the review a data licensing company runs when a supplier that has already licensed one dataset offers another. Company-level evidence, such as the entity, its signatories and a signed master agreement, can often be reused after a refresh check, while the rights in the new records, the new scope and the buyer’s fit test must be checked fresh.
How much cheaper is a second dataset from the same supplier?
It depends on where the qualification hours sit. In this article’s illustrative model a repeat deal costs 56.7 percent of a first when nothing has changed and 80.0 percent after a change of control, because dataset-level checks never shrink. Real figures need a company’s own matched cohort data.
Why does the overlap check get harder on a repeat deal?
If the first license was exclusive for AI training over the selected records, the second selection must not include them. Checking a new scope against everything already selected and delivered takes more work than scoping a first dataset, which is why the model gives it 9 hours on a repeat deal against 6 on the first.
What do web certificate rules have to do with data suppliers?
They are the clearest public example of rules for reusing checked evidence. The CA/Browser Forum lets identity checks be reused for 398 days but domain control only for 200 days now and 10 days from 2029, and it requires a record of what was verified and how. A data supplier ledger can follow the same logic.
Does SourceX publish repeat-supplier metrics?
No. Its live pages describe the path, a reused master agreement and a new Dataset Deal Confirmation for each deal, but not counts, hours or rates. Its investor page says company and commercial metrics are shared with investors directly, and materials can be requested there.
Sources
- The CA/Browser Forum Baseline Requirements (version 2.3.1) set maximum reuse periods of 398 days for identity validation and 200, 100 and 10 days for domain validation from 2026, 2027 and 2029, forbid reuse beyond those periods, require CAA processing within the record’s TTL or 8 hours, cap cached WHOIS data at 48 hours and, from 15 July 2026, require records of what was verified and the method used. Source
- Baseline Requirements version 1.7.4 (April 2021) allowed reuse of previous validations completed no more than 825 days before issuance. Source
- Baseline Requirements version 1.7.5 (ballot SC042) limited reused domain and IP address validation data to 398 days from 1 October 2021. Source
- Mozilla’s root store policy requires certificate authorities to obtain audits before inclusion and at least annually after, with successive period-of-time audits contiguous and without gaps. Source
- IAF MD 5 (2023) sets annual surveillance audit time at about one-third of initial certification audit time. Source
- IAF MD 5 (2019) sets recertification audit time at about two-thirds of the time an initial certification audit would need at the time of recertification. Source
- SourceX’s how-it-works page says one Master Supplier Licensing Agreement covers later deals, returning suppliers reuse it and sign a new Dataset Deal Confirmation, a material change needs a signed amendment, exclusivity covers only selected data, and suppliers confirm data has not been licensed for AI training before. Source
- SourceX’s investor page says one repeatable agreement shortens legal review for both sides on every deal and that company and commercial metrics are shared with investors directly. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.