GRIDS/365FYI
AI DECISION AUTHORITY ASSESSMENT · 01/04
01 · Bound

Bound the decision and the authority

Choose one consequential workflow. Define what AI may contribute, what must be evaluated, and who is actually authorized to act before discussing broader automation.

01

Write the authority contract

AI may propose for the decision to but only may authorize
Why this decision is consequential
Assessment boundary This aid helps a team expose responsibilities and design a bounded evaluation. It does not establish legal authority, determine compliance, validate a policy, certify an AI system, or prove that a proposed control is effective.
02

Assign the responsibilities

ResponsibilityMay doMust not doNamed ownerConfirmed?
AI assistant or agent
Governed model
Human reviewer
External authority / process
Discovery aid · Keep generated help, model evaluation, and recognized authority distinct.grids365.fyi · 1/4
GRIDS/365FYI
AI DECISION AUTHORITY ASSESSMENT · 02/04
02 · Trace

Trace one proposal through the system

Retain the AI contribution with attribution. Then connect it to approved facts, explicit model logic, an accountable decision, and a record that another reviewer can reconstruct.

01

Identify the AI contribution

02

Resolve the important facts

Fact used by the proposalApproved sourceRevisionAs-of timeStateOwner
03

Name the evaluation model

Rule, calculation, or constraintSource / authorityRevisionReason it appliesTested?
04

Follow the responsibility sequence

  1. AI proposal
  2. Approved facts
  3. Model evaluation
  4. Human decision
  5. Record / outcome
Keep unlike claims separate
Source factAssumptionAI proposalModeled findingRecommendationAuthorized decisionObserved outcome
A fluent recommendation is not a verified fact, approved rule, authorized decision, or observed outcome.grids365.fyi · 2/4
GRIDS/365FYI
AI DECISION AUTHORITY ASSESSMENT · 03/04
03 · Review

Design the review packet and safeguards

Human authority is meaningful only when the reviewer can inspect the basis, disagree with the proposal, use an escalation path, and act within a recognized scope.

01

Define what the reviewer must see

Review elementHow it will be shownEvidence sourceWho verifies itRequired?
Proposal and AI contribution
Facts, sources, and freshness
Rules, constraints, and model revision
Alternatives, reasons, and uncertainty
Authority, exception, and appeal path
Reviewer must be able to
Review conditions are real
02

Confirm safeguards outside the AI session

Do not treat process design as outcome evidence Attribution, validation, explanation, and human authorization can make a decision more inspectable. They do not prove that the source is complete, the policy is appropriate, the decision is lawful or fair, or the eventual outcome is good.
A reviewer needs context, recognized authority, and a practical way to disagree.grids365.fyi · 3/4
GRIDS/365FYI
AI DECISION AUTHORITY ASSESSMENT · 04/04
04 · Evaluate

Run a bounded evaluation and decide readiness

Test one useful slice under controlled conditions. Record the baseline before choosing targets, keep external authority in place, and decide whether to proceed, narrow, prepare, or stop.

01

Write the evaluation statement

For allow AI to then evaluate with while retaining
02

Choose measures before observing results

MeasureCurrent baselineObservation methodDesired criterionObserved result
Correctness against approved cases
Unsupported or stale facts detected
Reviewer can reconstruct the finding
Human correction / deferral / rejection behavior
Workflow-specific safety or quality measure
03

Score readiness

DimensionReadyNeeds workNot yetEvidence or next action
Decision and authority boundary
AI contribution and attribution
Approved facts, rules, and acceptance cases
Reviewer capability, recourse, and safeguards
Evidence record and outcome observation

Bring this completed assessment to a bounded Grid evaluation.
Bring the real workflow, source and policy revisions, representative cases, existing review process, and the person who owns the decision. A completed form is a discovery input—not approval to automate.

Continue at Grid Dev