Decentralised News · Agentic Finance · Decision research
By Heath Muchena · Source review 2026-11-11 · Instrument version 1.0.0
What Matters
An answer service sells a sequence of disclosures. The commercial question is how much useful information a permitted buyer can retain and combine across that sequence. DN’s synthetic model counts directly returned records; the six checks also cover aggregation, narrow populations, versions, joins and retention. No live vendor, private dataset or customer behaviour was tested.
A supplier can restrict each response to a few facts and still deliver a substantial reusable collection over a contract’s lifetime. That outcome may be exactly what the buyer paid for. Trouble starts when the supplier prices isolated answers while its licence, product design and renewal assumptions depend on buyers never assembling them.
The right starting point is a defined commercial asset: which records, fields, versions and permitted uses make this dataset valuable? Only then can a publisher distinguish legitimate cumulative use from a request that requires broader rights. The instrument below examines a deliberately narrow case before the six checks extend the decision.
DN Evidence Block
Sources reviewed: 2026-11-11 · Evidence type: Primary-source synthesis, an implemented synthetic set-union model and labelled hypothetical examples.
Scope: Commercial answer services and cumulative information disclosure. The calculator counts directly named IDs in a fixed synthetic dataset; the article identifies additional questions outside that calculation.
- NIST SP 800-226, final March 2025, explains reconstruction, auxiliary information and composition of differential-privacy guarantees. DN’s counter does not implement those guarantees.
- Datasheets for Datasets, version 8, 1 December 2021, provides a documented-dataset framework covering composition, uses, distribution and maintenance.
- W3C Data on the Web Best Practices, supplies publication guidance for provenance, versions, licences and controlled sharing.
- Synthetic example: three answer arrays contain nine record mentions and seven distinct IDs in a twelve-record universe. These are model inputs, not observed buyer behaviour.
Not established by this work: No live API queries, real customer reconstruction, access-control bypass, private personal data, vendor revenue impact or formal privacy guarantee.
Original DN instrument · version 1.0.0
DN Synthetic Reconstruction Model
Count distinct synthetic records across answer arrays. Each integer represents one complete fictional record in a fixed dataset. Do not enter private customer data or real credentials.
Enter your assumptions or load a labelled example. Required blank numbers stay missing; unknown evidence remains unknown. Use nonpersonal labels only. This instrument makes no network requests and uses no persistent browser storage.
Use your own inputs
No current result. Complete inputs or load an illustrative example.
A set-union counter for directly named synthetic records. It does not estimate aggregate inference, field completeness, external joins, changing versions, privacy loss, licence compliance or commercial damage. All user entries remain unverified.
1. Direct answers accumulate into a reusable collection
The most straightforward route requires no inference. A buyer receives useful records, keeps them within the agreed rights and combines subsequent answers with the earlier ones. Limiting the size of one response does not define the size of the accumulated collection.
Synthetic example: a fictional catalogue contains twelve complete records. The first answer returns IDs 1, 2 and 3; the second returns 3, 4 and 5; the third returns 5, 6 and 7. There are nine mentions, seven distinct records and two repeated mentions. Direct coverage is 7 ÷ 12, or 58.33%. Three answers have delivered more than half the catalogue in this deliberately simple case.
The denominator matters. A record can contain an identifier, a price, a licence and a dated availability claim. Receiving its identifier alone is not receiving that complete record. The tool assumes each listed ID represents a whole defined synthetic record; a real assessment must state which fields make a record commercially useful.
Decision check: price and license the intended cumulative use. If a customer needs a maintained reference collection, compare a dataset licence with repeated answer purchases instead of assuming a small response prevents collection building.
2. Aggregate answers can reveal relationships between records
A response containing totals can disclose useful relationships without listing individual records. In a fictional stock report, one approved total for a set and another for a known subset may reveal the difference between them. Whether that difference identifies a record depends on the surrounding information and the release design.
NIST SP 800-226 explains that sufficiently informative statistical releases can support reconstruction and that aggregation alone does not establish a privacy guarantee. This is a documented risk category, not a finding that any named DN supplier is vulnerable.
For product design, ask what combinations the service is intended to allow. A price-index product, a row-level supplier database and a one-off procurement brief have different utility requirements. Removing every useful relationship would defeat some of those products; treating every aggregate as harmless would miss their cumulative disclosure.
Decision check: review the release family together, using authorised synthetic fixtures. Record which relationships are deliberately disclosed, which are outside the permitted product and who owns that decision. The direct-ID tool above cannot quantify this form of inference.
3. Narrow filters make an answer more specific than its label suggests
A question described as a market summary may concern a very small market. Region, service type, capacity, date and contract size can narrow an answer until its practical meaning is one supplier or one commercially sensitive event. The headline “aggregate” says little about that final specificity.
Hypothetical fixture: a fictional catalogue contains many delivery providers, but only one is listed for a particular route and equipment class on a specified date. An answer about “providers in this category” may therefore describe that provider even if the output omits its name. The example establishes a logic problem, not the prevalence of such cases in a real service.
A useful review captures the complete filter combination and the population it refers to. Testing each filter independently would miss the narrowing produced by their intersection. The reference population must also be dated: a category with several participants today might have had only one last month.
Decision check: include small-population cases in the approved release specification. Determine when to generalise a response, withhold a detail or offer an explicitly licensed record. A universal minimum group size would require evidence this article has not collected.
4. Repeated answers reveal a history that one snapshot does not contain
A buyer can learn from change as well as from coverage. Retaining a permitted answer about a fictional offer in January and another in February creates a two-point history. That history may help compare price movements, service withdrawals or changed conditions even if the supplier sells only a “current answer”.
The useful unit is therefore sometimes a record-version pair. If ID 7 appears in three dated releases with different terms, a simple unique-ID count remains one while the buyer’s collection becomes richer. Our calculator intentionally holds the dataset fixed and cannot measure this additional information.
Distinguish publication date, effective date and observation date. A February retrieval of a January offer is not necessarily evidence of a February change. Corrections also need their own label: silently replacing a mistaken value can make a customer infer a business event that never happened.
Decision check: state what the subscription permits buyers to retain, how corrections reach retained copies and whether historical versions form part of the paid product. Test the history with fictional revisions before drawing conclusions about actual customers.
5. Answers become more revealing when combined with outside context
A response does not arrive in an empty room. Buyers may already have public catalogues, their own transactions, prior research and openly published company information. A modest answer can become valuable when it supplies a missing link between those records.
NIST’s guidance discusses auxiliary information in the context of linking and privacy. The commercial extension here is DN’s analysis: review what an authorised buyer can combine, rather than assessing every response as an isolated object. That does not turn public information into a contractual right to use every other input.
Hypothetical fixture: an answer names a product family and delivery month. A permitted public catalogue supplies a compatible model mapping. Together they answer a procurement question neither source settled alone. The combination may be the intended benefit of the product; it should not automatically be classified as abuse.
Decision check: document foreseeable joins and their rights. Use a controlled synthetic or licensed reference set for testing, and preserve the origin of each field. The counter’s “IDs absent from this log” result must never be presented as “information the buyer cannot know”.
6. Retained outputs outlast query limits and access changes
A query allowance controls future service use. It does not, by itself, describe what happens to answers already delivered. A customer may have saved them in a report, decision log or licensed internal reference collection. Closing an account does not demonstrate that those copies disappeared.
This creates a pricing and continuity choice. A supplier can offer narrowly scoped answers, a time-limited access service, a retained snapshot or a broader maintained dataset. Each has a different promise. The terms should say which promise applies, including permitted sharing and the treatment of previously delivered material after termination.
Counting calls alone also hides reuse. One answer may support many internal decisions, while several calls may repeat the same records. In the synthetic example, nine mentions buy seven distinct records. Neither number establishes the customer’s financial gain or the supplier’s lost revenue.
Decision check: maintain a release-and-rights record alongside usage billing. Review renewal value against the continuing service—new coverage, verified updates or support—not an assumption that earlier answers lose all usefulness. Where persistent access is intended, price it directly and make the permitted retention intelligible.
Methodology, limitations and corrections
Method: the six checks separate disclosure mechanisms that require different evidence. DN reviewed the cited primary documents and implemented the direct-ID counter with labelled fictional data. The formula is distinct IDs in all answer arrays ÷ fixed dataset size. Repeated mentions equal total mentions minus distinct IDs. These quantities describe the supplied input only.
Worked result: the twelve-record example returns three answer events, nine mentions, seven distinct IDs, two repeats, 58.33% direct coverage and five IDs absent from the log. Repeating the same three-record answer three times produces nine mentions but only three distinct IDs, or 25% coverage. Neither result measures the usefulness of those records.
Evidence design: follow Datasheets for Datasets by stating what the dataset contains and its intended uses. Use W3C publication guidance to keep provenance, version and licensing information alongside the outputs. This article does not claim conformance to a dataset standard.
Boundary: NIST’s formal treatment of repeated differentially private releases concerns composed privacy guarantees. DN’s percentage is not such a guarantee and cannot be converted into one. A source-based explanation of a risk is also not an observed result from DN’s tool.
Practical next measurement: a supplier can review authorised synthetic answer sequences against a predefined release policy, recording exact fields and versions disclosed, legitimate tasks completed and unresolved conditions. Set the commercial objective before choosing a control. Reduced disclosure can also reduce the usefulness customers are buying.
For corrections, identify the claim, source and date when contacting Decentralised News. A new record-creation timestamp does not refresh its underlying evidence.
Change log: version1.0.0, 2026-11-11. Initial documentation review, original instrument and labelled examples.
Frequently asked questions
Does a high direct-coverage percentage prove a licence violation?
No. Broad accumulation may be authorised. The percentage counts distinct synthetic IDs in the supplied answers; the actual licence, intended use and facts determine what rights need review.
Is this a differential-privacy calculator?
No. It does not implement a privacy mechanism or compose privacy parameters. It counts directly named synthetic records and omits many forms of inference.
Can an empty answer reveal information?
Possibly. The absence of qualifying records can itself be informative. An empty array adds no direct IDs in this model, which is narrower than measuring everything an answer communicates.
Should answer products always become dataset subscriptions?
No. Compare the customer’s defined job, permitted retention, ongoing update needs and the cost of repeated answers. Some decisions need a bounded answer; others need a maintained collection.
Primary sources
- NIST SP 800-226: Guidelines for Evaluating Differential Privacy Guarantees, March 2025. Reconstruction and auxiliary-information risks; cumulative privacy guarantees. The DN counter does not implement differential privacy. Reviewed 2026-11-11.
- Datasheets for Datasets, arXiv version 8, 1 December 2021. Documenting dataset composition, intended uses, distribution and maintenance. Reviewed 2026-11-11.
- W3C Data on the Web Best Practices. Provenance, versions, licences and the need to consider controlled sharing. Reviewed 2026-11-11.
Turn the finding into a decision
One decision for the DN Research Desk: should a defined customer job be served by individual answers, retained snapshots or a broader dataset licence? A scoped brief can compare the permitted use, cumulative disclosure and update obligations using an authorised synthetic fixture pack. Submit the decision and available evidence to the Research Desk. Scope, deliverables, price and access permission must be agreed before work begins.
Related DN research
Disclosure: Analysis and instrument results are editorial. Where labelled affiliate links appear, DN may receive compensation when you sign up or buy through them. Commercial relationships do not determine findings. No tested integration or guaranteed outcome is implied.
