Practical guide
What evidence shows records retention is operating?
Records-retention control is operating only when an organization can trace covered records from an approved schedule through classification, preservation, cutoff, disposition eligibility, hold checks, authorization, execution, and exception review. A policy or configured retention period is design evidence, not operating evidence. The strongest test selects real record classes and reconstructs.
Records-retention control is operating only when an organization can trace covered records from an approved schedule through classification, preservation, cutoff, disposition eligibility, hold checks, authorization, execution, and exception review. A policy or configured retention period is design evidence, not operating evidence. The strongest test selects real record classes and reconstructs what happened, including records correctly retained, transferred, deleted, suspended from disposition, or held as exceptions.
Define the control objective and boundary
Use a bounded objective such as: “Covered AI-enabled-work records are identified, retained, protected, and disposed of under approved instructions, while holds and exceptions prevent premature disposition and completed actions remain reproducible.”
Map the systems, repositories, record classes, business owners, records owner, legal or regulatory reviewers, custodians, data locations, and service providers in scope. Include prompts, outputs, source packets, approvals, logs, assessment evidence, or model records only where the approved schedule and purpose require them. Do not retain everything by default merely because storage is available.
NARA guidance for U.S. federal records says disposition instructions should cover inventoried and evaluated records, use clear cutoffs and retention periods, and avoid ambiguous directions such as “when no longer needed.” It is federal records-management guidance, not a universal retention rule, but it provides a primary-source example of precise, controllable disposition instructions. NARA: Preparing disposition instructions
Sources: [1]
Use the CHAIN evidence test
CHAIN distinguishes a documented design from evidence that the control ran.
### C: Catalogue and schedule evidence
Verify that an approved schedule maps each in-scope record class to its owner, authoritative repository, cutoff event, retention period, final disposition, and applicable authority. Sample system inventories, data maps, contracts, and workflow documentation to find record locations missing from the schedule.
Useful artifacts include the approved schedule, record-class inventory, system map, source authority, version history, and approval record. A schedule that omits a live repository is a design gap even if the configured systems behave consistently.
### H: Hold and exception evidence
Test how legal holds, investigations, complaints, preservation duties, incidents, access requests, and approved exceptions interrupt ordinary disposition. Select active and released holds and trace the affected record classes, custodians, systems, dates, acknowledgements, and release authority.
Look for a hold register, exception register, custodian notices, system suspension logs, reconciliation results, and expired-exception review. Confirm that a released hold returns records to the correct schedule rather than causing immediate uncontrolled deletion.
### A: Application evidence
Compare the approved schedule with actual system configuration and repository behavior. Inspect retention rules, labels, event triggers, default paths, backups, exports, archives, and vendor settings. Confirm that users cannot silently move covered records into an unmanaged route.
Configuration screenshots are weak on their own. Prefer machine-readable exports, change history, access logs, and a sampled calculation of eligibility dates. Reperform the date logic for fixed and event-based rules. Record timezone, event source, version, and any dependency that can delay or alter the calculation.
### I: Instruction-to-action evidence
Sample in both directions. From the schedule, select record classes and find actual records. From repositories, select records and identify the controlling schedule. Include records expected to remain, records eligible for disposition, records already disposed, permanent records transferred where relevant, and records suspended under hold.
For each completed disposition, verify the eligibility calculation, hold check, authorization, execution result, scope, timestamp, operator or service identity, and reconciliation. For retained records, verify availability, integrity, access, and the next relevant event. Do not infer successful deletion merely from absence in the primary interface if backups, replicas, or exports remain in scope.
### N: Notice, review, and next action
Review failed jobs, unmatched records, expired holds, configuration drift, vendor changes, restored backups, inaccessible archives, unauthorized deletion attempts, and overdue transfers. Each exception needs an owner, reason, affected scope, compensating action, expiry, and review date.
The NARA Records Schedule illustrates that retention instructions tell staff how to maintain operational records and what final action to take, including transfer of records designated permanent. An organization must use the authority and schedule that actually apply to it. NARA Records Schedule
Sources: [2]
Minimum evidence packet
| Evidence layer | Example artifact | Operating question | | --- | --- | --- | | authority | approved schedule and version | Is there a current instruction for this class? | | coverage | inventory and repository map | Are all actual record locations represented? | | configuration | rule export and change history | Does the system implement the approved instruction? | | eligibility | sampled cutoff calculation | Was the correct record eligible at the correct time? | | hold | hold match and release log | Was disposition suspended and resumed correctly? | | authorization | approved batch or transfer record | Did an authorized owner approve the action? | | execution | immutable job result and reconciliation | Did the intended action complete across scoped locations? | | exception | failure and remediation register | Were gaps owned, bounded, and reviewed? | | monitoring | periodic control review | Did drift or overdue action trigger a disposition? |
Retain enough evidence to reproduce the control decision without keeping unnecessary content from the underlying records. The evidence packet itself needs an approved retention rule and access boundary.
Worked example: AI-assisted support records
An organization’s policy says support interaction records are retained for three years after case closure. The application shows a three-year setting, and the control owner reports no failures. CHAIN exposes whether that statement is supported.
Catalogue: the approved schedule covers case messages and final resolutions, but the system inventory also finds AI draft suggestions in an analytics export and attachments copied to a separate object store. Neither location is mapped. The control design is incomplete until the records owner classifies those artifacts and updates the schedule or prevents the copies.
Hold: one customer dispute placed a case and attachments on hold. The primary case store was suspended correctly, but the analytics export did not receive the hold identifier. The organization pauses disposition for the affected export scope and opens a remediation item.
Application: the configured three-year rule uses case creation rather than closure. A sample case reopened after two months would become eligible too early. The owner repairs the event source, documents the configuration change, and recalculates the affected population.
Instruction to action: a repaired sample includes ten retained cases, five eligible cases, two completed deletion records, and the held dispute. The reviewer reproduces cutoff dates, confirms authorization, verifies deletion across the primary store and scoped replicas, and reconciles the deletion job count. Backup handling remains under a separate approved rule and is stated rather than assumed.
Notice: the control disposition is not operating as designed; restrict and repair. Required evidence for closure includes the updated inventory and schedule decision, hold propagation test, corrected event rule, affected-population review, and a clean disposition sample. The original policy and configuration screenshot are retained as evidence of the detected gap, not overwritten by the repair.
Failure and hold conditions
Hold an effectiveness conclusion when a live repository is unmapped, schedule authority is unclear, disposition dates cannot be reproduced, holds do not reach every scoped location, deletion or transfer lacks authorization, execution is not reconciled, exceptions do not expire, or restored data bypasses the rule. Also hold when jurisdiction, contract, litigation, privacy, or sector requirements need current specialist review.
Two Prune can use records-retention review as a realistic assessment scenario and describe task-bounded evidence across framing, evidence navigation, AI use, verification, judgment, and communication. It does not certify records compliance, provide legal advice, or make employment decisions.
Sources
U.S. National Archives: Preparing disposition instructions
- U.S. National Archives: NARA Records Schedule
Sources: [1] [2]
Sources
- 1.NARA: Preparing disposition instructions · U.S. National Archives
- 2.NARA Records Schedule · U.S. National Archives