Practical guide
What evidence shows a privacy control is operating?
Evidence that a privacy control is operating must connect a defined privacy objective to the actual process, population, period, test, exceptions, and corrective action. A policy, screenshot, training completion rate, or system setting may show design or implementation, but none alone proves that a control operated as intended across the relevant activity.
Evidence that a privacy control is operating must connect a defined privacy objective to the actual process, population, period, test, exceptions, and corrective action. A policy, screenshot, training completion rate, or system setting may show design or implementation, but none alone proves that a control operated as intended across the relevant activity.
Source, legal, and assurance boundary
NIST SP 800-53A provides customizable procedures for assessing security and privacy controls and analyzing results. The NIST Risk Management Framework describes the assessment purpose as determining whether controls are implemented correctly, operating as intended, and producing the desired outcome. The UK Information Commissioner’s Office says accountability measures are ongoing and organizations should keep evidence, maintain processing records, apply appropriate measures, record breaches, and review arrangements. These sources support an evidence method. They do not establish legal compliance in a specific jurisdiction or certify any control described here. NIST: SP 800-53A Rev. 5 NIST: RMF Assess Step ICO: Guide to accountability and governance
Sources: [1] [2] [3]
Build an EVIDENCE chain
### E — Establish the objective and boundary
State the privacy outcome in observable terms. Identify the processing activity, personal-data categories, people affected, systems, locations, vendors, channels, period, policy or requirement, risk, control owner, operator, reviewer, frequency, and trigger for reassessment.
Avoid objectives such as “protect privacy.” A testable objective is: “Only approved fields from the support platform enter the summarization workflow, and prohibited identifiers are blocked and recorded before transfer.”
### V — Verify control design
Inspect whether the documented procedure could achieve the objective. Check input rules, roles, access, approvals, configuration, logging, exception authority, fallback, retention, incident route, and change management. Trace each design statement to a current artifact and version.
Design evidence can include a data-flow map, processing record, risk assessment, approved procedure, configuration baseline, vendor terms, test specification, role assignment, and change record. It supports only a design conclusion until operation is tested.
### I — Inspect implementation
Confirm that the design exists in the operating environment. Observe the workflow, inspect configuration directly, interview operators, and examine a transaction from intake to retained evidence. Use more than inquiry when consequences are material.
Record environment, version, date, assessor, access limitations, and deviations from the approved design. A screenshot without provenance or a demonstration in a test tenant may not establish production implementation.
### D — Define the population and test
Identify every eligible event in the period and reconcile it to the system of record. Define inclusion, exclusion, duplicate, failed-event, retry, and missing-data rules. If sampling, document the selection method and why it is suitable for the decision.
For each item, define the evidence expected and the procedure: inspect, observe, inquire, or reperformance. Reperform the control where feasible. Preserve event IDs and hashes or other references without copying unnecessary personal data into the assessment record.
### E — Examine exceptions and counter-evidence
Classify failures by objective, cause, affected population, duration, detectability, consequence, and recurrence. Include events caught by the control, events missed by it, overrides, disabled periods, manual workarounds, monitoring gaps, incidents, complaints, and stale exceptions.
Do not erase a failure because it was later corrected. Record containment, notification decisions, root-cause work, corrective action, retest, and residual risk. Distinguish a control detecting a prohibited event from a control preventing it.
### N — Narrow the conclusion
Conclude only for the tested objective, implementation, population, period, and procedure. Useful dispositions include design not established, implemented but not tested, operating for tested items, operating with exceptions, not operating as intended, and insufficient evidence.
State coverage, sample, exceptions, missing evidence, assessor independence, limitations, and review date. Do not convert absence of detected incidents into evidence that the control is effective.
### C — Correct and confirm
Assign corrective actions with owners, due dates, acceptance evidence, and escalation. Retest the changed control and affected routes. Confirm that the evidence record and processing documentation reflect the new version.
### E — Expire and reassess
Set an evidence expiry based on consequence and change frequency. Reassess after material changes to data, purpose, system, model, vendor, access, workflow, jurisdiction, incident pattern, or control design. Historical evidence remains evidence of the historical state, not the current one.
Control evidence record
| Field | Evidence to retain | | --- | --- | | objective and scope | exact outcome, processing activity, population, period | | ownership | accountable owner, operator, reviewer, escalation authority | | design | procedure, configuration baseline, data flow, exception route | | implementation | observation notes, environment, version, approved settings | | test | population reconciliation, sample, procedures, results, assessor | | exceptions | affected events, cause, consequence, containment, residual risk | | correction | action, owner, due date, retest, closure evidence | | conclusion | narrow disposition, limitations, review and expiry dates |
Worked evidence review: synthetic AI research assistant
An internal research assistant summarizes approved interview notes. The privacy control is intended to prevent direct identifiers from entering the AI workspace. The procedure requires a preprocessing service to remove name, email, phone, and account-number fields, blocks transfer when a prohibited pattern remains, and logs the source record, rule version, result, reviewer, and override.
The assessor defines the period as April and the eligible population as 1,284 attempted transfers. The event log reconciles to 1,279 completed or blocked events; five retries have no final status. This is a population gap, not five compliant events.
Design inspection finds current field rules, an approved override route, and a weekly exception review. Implementation inspection finds that the production service uses the approved rules, but one regional import maps a free-text “contact” field outside the standard schema. The assessor reperforms 40 risk-weighted events and inspects all 17 blocks and three overrides.
Thirty-nine sampled events contain no prohibited direct identifier after preprocessing. One free-text record retains a phone number because the regional mapping bypassed the standard field rule. The control did not block it. The review also finds that two overrides have approval but no reason code and that the weekly review missed the five unresolved retries.
The disposition is operating with material exceptions for the tested April population. The system blocks known fields, but the regional path, unresolved retries, and incomplete override evidence prevent a broader conclusion. The owner disables the regional route, adds schema and pattern tests, resolves the five events, requires override reasons, and schedules a full-population reconciliation after release.
The retest verifies the repaired route and all April exceptions. It does not establish legal compliance, absence of other identifiers, or operation in later months. A jurisdiction-specific privacy and legal review remains separate.
Using assessment evidence appropriately
A realistic work simulation can show whether a participant defines a privacy objective, reconciles a population, tests evidence, identifies exceptions, and narrows a conclusion. The simulation is not an audit of the participant’s employer or proof that a real control works.
Two Prune provides evidence about observed participant performance in bounded assessments. It does not certify privacy controls, provide legal advice, determine regulatory compliance, establish control effectiveness outside the tested case, or make employment decisions.
Sources
NIST: SP 800-53A Rev. 5
- NIST: RMF Assess Step
- ICO: Guide to accountability and governance
Sources: [1] [2] [3]
Sources
- 1.NIST: SP 800-53A Rev. 5 · National Institute of Standards and Technology
- 2.NIST: RMF Assess Step · National Institute of Standards and Technology
- 3.ICO: Guide to accountability and governance · Information Commissioner's Office