Match count-noise scale to the protected contribution

Translate the proposed promise into a review question

A fictional arts publisher wants a public summary of participation in an online writing activity. The communications draft says the report protects participants, while the technical note discusses individual submissions. One participant may submit more than once. The editor should not resolve that mismatch by choosing whichever wording sounds more reassuring. The privacy owner needs a clear account of the promised protection and the intended report.

Keep the proposed report separate from the publication draft. Identify its audience, purpose and the person responsible for privacy review. Ask whether the organization has a legitimate reason and appropriate authorization to use the underlying information for this purpose. A technical exercise does not answer those governance questions, and a compelling communications deadline does not make them disappear.

Stipulate each person contributes an integer from 0 through 5 to one total f. Neighbors add or remove one person’s entire contribution. The largest total change is 5. For ε=1, NIST’s Laplace rule uses scale b=5/1=5, not 1.

The numbers are invented and do not describe any actual participant. They are a compact specification for discussing the mismatch. Do not retrieve real submissions to make the example more convincing. The editor can identify the question that requires expert review without receiving personal records or building a new reporting tool.

Inspect the complete bounded comparison

For neighboring contributions [2,5] and [2], totals are 7 and 2. Release f+Z only in this mathematical model, with density p(z)=exp(−|z|/b)/(2b). For output interval [7,8], probabilities are (1−exp(−1/b))/2 and (exp(−5/b)−exp(−6/b))/2; their ratio is exp(5/b).

At b=1 the ratio is exp(5), exceeding the required exp(1) bound. At b=5 it is exp(1). The general guarantee also relies on the declared global contribution bound. This is a single ideal mechanism, not software certification, repeat-release accounting or a claim of anonymity.

Keep the full input definition beside the comparison. A report that presents only the selected parameter leaves another reviewer unable to tell which assumptions justified it. The local verification evaluates the stated formulas and an independently expressed probability calculation. It does not create a release, sample real data or establish that any vendor’s software implements the intended model.

Give the privacy owner the proposed public sentence and the technical specification together. Ask them to identify the exact claim they can support, the unresolved assumptions and the evidence still required. Do not turn the teaching result into a recommended organizational setting. Choosing a real approach requires judgment about the particular purpose, context and risk.

If the owner requests a change to the report, preserve the reason in the editorial record. A smaller or differently described publication may be more appropriate than the original plan. The communications team should understand what has changed so it does not quietly restore an unsupported sentence when polishing the announcement.

Separate analytical usefulness from permission to publish

Ask what decision the proposed summary is meant to support. The publisher may want to describe participation, decide whether to run another activity or thank the community. Those purposes do not necessarily need the same public statistic. A narrative about the activity, an approved aggregate or a statement that omits the disputed number may serve the editorial job differently. Compare them honestly rather than treating the original chart as indispensable.

Have the relevant owner explain the limitations of any approved output in language the editor can preserve. Do not remove a qualification because it makes a caption longer. If the output cannot support the proposed comparison, revise the claim rather than asking the analyst to produce a more convenient number. The publisher’s desire for a clear story should not dictate what the evidence is allowed to mean.

Keep technical review, content approval and release permission as separate recorded decisions. A colleague can agree that the mathematics is internally consistent without authorizing a public report. Likewise, an editor can approve the wording while waiting for the required privacy decision. Make that incomplete state visible so a scheduled publication does not proceed on the strength of an ambiguous approval message.

Close with a claim that has an accountable owner

Before publication, reconcile the selected artifact with the approved public wording and the intended audience. Check that no unreviewed appendix, downloadable worksheet or explanatory note has been added to the package. The review should concern what will actually be shared, not an earlier draft that omitted supporting material. Any material scope change should return to the responsible owner.

If review remains unresolved, state that the report is not ready and choose any separately authorized alternative. Do not describe a postponed chart as a technical failure or imply that a privacy label alone settles the question. A careful delay can preserve the organization’s ability to explain its choice accurately without promising a protection it has not established.

Keep the final decision concise: what was approved, for which purpose and audience, which limitations must remain visible and who owns future changes. The synthetic calculation belongs in the training record, not as a substitute for that approval. No personal information was accessed or transmitted, no security setting was changed and no public reporting system was operated in this lesson.

Make the reasoning concrete.

The arts publisher has draft wording promising protection for participants, but the accompanying technical note has not been reconciled with that promise.

The editor holds the unresolved claim and requests a bounded review without accessing participant records.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Promise

   What does the proposed public sentence say is protected?

2. Purpose

   Why does the audience need this report?

3. Specification

   What complete assumptions accompany the proposed calculation?

4. Authority

   Who can approve the actual artifact and its audience?

5. Change

   Which revisions require renewed review?

Watch for the wrong conclusion.

• Treating technical consistency as publication permission.
• Polishing away a required limitation.
• Using a synthetic calculation to certify an unexamined product.

Sources & scope

Original fictional reporting scenario and synthetic probability comparison. The source supports the defined mathematical boundary. This is educational analysis, not a privacy audit, legal advice or authorization to disclose information.

• NIST SP 800-226: Guidelines for Evaluating Differential Privacy Guarantees →Primary government guidance · Supports: Supports the privacy unit, sensitivity and Laplace scaling.