StackPilot / Occupation guides / Manual 024

Occupation manual 024 · review-only workflow

AI can organize document-control evidence. Records authority stays human.

Use AI to preserve approved source labels, owners, versions, and unknowns. Do not let it access a repository, decide record status or retention, change metadata or permissions, delete or destroy documents, publish, or take external action.

For: document-control and records teamsOutput: source & version control packetReviewed: 2026-07-30
Answer-first field map

Build a review packet, not a records engine.

Document Management Specialists administer systems and procedures that capture, store, retrieve, share, and destroy electronic documents and records.1 That scope makes provenance useful, but also makes silent inference risky. A safe AI workflow prepares a labeled packet from fictional or organization-approved minimum-necessary text. A named human decides every record, source, retention, access, and system question.

InputApproved labels only: source, owner, date/version, stated use, and known gaps.
OutputA Source & Version Control Packet ending in needs_review.
Not an outputNo repository search, record classification, retention rule, hold, metadata edit, permission change, publication, deletion, or message.
Federal-source limit: NARA guidance explains federal records practices. It is useful context, not a universal record definition, retention instruction, or private-organization policy.
  1. Set the boundary before any AI work.

    Give the packet a fictional or approved scenario label, purpose, and authority owner. Exclude live document bodies, repository exports, personal or confidential data, legal-hold material, schedules, credentials, and system identifiers.

  2. Register what is known without selecting a winner.

    For each label, record the stated source, owner, date/version, intended use, and status: approved, unknown, or conflicting. Never ask the model to decide which version controls.

  3. Ask AI to organize, not retrieve or classify.

    AI may format the register, find missing fields in the supplied labels, group duplicate labels, and draft neutral questions. It must not connect to a repository, search unapproved content, infer a classification, or create metadata.

  4. Turn gaps into reviewer questions.

    For every missing owner, dated version, access condition, retention rule, hold, or approval, preserve the gap as unknown. Ask a named authority which current approved rule applies.

  5. Route authority to the right owner.

    Records/information governance owns record and retention questions; business/data owners own purpose; legal owns hold and legal questions; privacy/security owns permitted handling; system owners own access and configuration.

  6. Stop at the handoff.

    Issue a local needs_review receipt. Do not open a system, change a record, publish, share, delete, destroy, transfer, sign, or communicate.

Narrow agent roles

Three bounded roles. One human control point.

Label stewardNormalizes supplied source labels and flags blank owner, date, version, or stated-use fields. No retrieval or source selection.
Conflict clerkGroups duplicated or contradictory labels and marks them unknown or conflicting. No record, retention, or authority conclusion.
Packet assemblerFormats neutral reviewer questions and a handoff receipt from the approved labels. No system connection, metadata, access, or external output.

NIST’s Generative AI Profile is voluntary, cross-sector risk guidance. Use it as a governance lens, not as a records policy or approval.5

Safe versus approval-required

Let AI prepare evidence. Let authority control records.

AI may prepareNamed human approval is required
Format labels supplied in the approved scope.Access, search, retrieve, or upload from a repository.
Preserve source, owner, version/date, stated use, and unknown/conflict labels.Select a source of truth, determine record status, classify, or set retention/disposition.
Draft neutral questions about missing authority or conflicting labels.Place/release a hold; delete, destroy, transfer, publish, sign, disclose, or share.
Produce a local review-only receipt.Change metadata, permission, system configuration, vendor/tool status, or any external record.
Stop immediately if the request requires a live document, repository, system ID, personal/confidential content, retention schedule, legal-hold matter, access decision, or an action affecting a record, person, system, or external party.
Failure modes

Five ways document control quietly turns into unauthorized control.

MistakeWhy it failsRepair
“Use the newest file.”Newest is not necessarily approved, authoritative, or applicable.Preserve dates and versions; ask the owner which source controls.
“Classify this for me.”Classification can create records, legal, access, or retention consequences.Record the question and route it to the authorized records owner.
“Clean up this folder.”Moving, deleting, or renaming can alter evidence, access, or disposition.Use a fictional label register; obtain separately scoped system approval for any implementation.
“The policy says keep it seven years.”The policy may be stale, noncontrolling, contextual, or subject to a hold.Label the rule/source/version unknown until an authorized reviewer confirms it.
“Send the final version.”Sharing or publishing can be an external action with privacy, contract, or legal effects.Stop at a local review receipt; human owners approve content, audience, and channel.
Runnable local artifact

Document Control Evidence Desk

Enter only fictional or organization-approved, minimum-necessary labels. This form stays in the browser and creates a review-only receipt; it does not send or save anything.

Primary sources

What this manual is built on.

Sources are evidence context, not permission to use private material or take records/system action. Recheck them and the controlling organization rules before any non-fictional use.

Related guides

Where this guide fits.

Proposed links: the Data Warehousing Specialist Evidence Desk for data-provenance work, the Database Administrator Change Evidence Desk for controlled database changes, the Web Administrator Operations Manual for site operations, and the Occupation AI Workflow Guide Directory for the broader shelf. Use these related guides to keep each workflow's authority boundary clear.