Override Justification Sufficiency Evaluation Prompt

This prompt is for audit leads, compliance engineers, and governance teams who need to programmatically determine if a single human override justification contains enough detail to stand alone in an audit. The job-to-be-done is sufficiency gatekeeping: before an override record enters a permanent audit trail or is accepted as a valid exception, you need a consistent, automated check that the justification meets your organization's minimum evidentiary standard. The prompt evaluates one justification at a time against a defined sufficiency rubric and returns a pass/fail assessment with specific gaps identified, enabling downstream routing logic to either accept the record or send it back to the reviewer for more detail.

Use this prompt when you have a defined sufficiency policy—a written standard for what constitutes an adequate override justification in your organization. Typical criteria include: the override reason is stated explicitly, the risk of not overriding is acknowledged, alternatives considered are documented, and the reviewer's identity and authority are captured. The prompt works best as part of a structured override pipeline where justifications are already captured in a machine-readable format (such as the output of an Override Justification Structured Log Prompt). Do not use this prompt as the sole gate for regulatory submissions without human review; it is a triage and quality assurance tool, not a legal determination.

This prompt is not a replacement for policy creation, nor does it evaluate whether the override decision itself was correct. It only checks whether the justification record is sufficiently detailed. Avoid using it when your sufficiency criteria are vague, contradictory, or still under development—the model will amplify that ambiguity. If your override workflow involves multiple stakeholders, combine this prompt with a Multi-Stakeholder Approval Routing Prompt upstream and an Override Audit Trail Completeness Verification Prompt downstream. For borderline scores, always route to a human reviewer rather than making an automated final decision.

The following prompt template is designed to be copied directly into your AI harness. It evaluates a single override justification against a defined set of sufficiency criteria and returns a structured pass/fail assessment with specific gaps identified. The template uses square-bracket placeholders for all dynamic inputs, making it ready for programmatic substitution in your application code. Before using this prompt, ensure you have defined your organization's minimum sufficiency criteria—these are the non-negotiable elements that every override justification must contain to be considered complete.

textCopy
You are an audit quality reviewer evaluating the sufficiency of a human override justification. Your task is to determine whether the justification contains enough detail to stand alone in an audit without requiring the reviewer to seek additional context.

## INPUTS

**Override Justification:**
[OVERRIDE_JUSTIFICATION_TEXT]

**Original AI Decision Context:**
[ORIGINAL_AI_DECISION_SUMMARY]

**Minimum Sufficiency Criteria:**
[SUFFICIENCY_CRITERIA]

**Risk Level of Decision:**
[RISK_LEVEL]

**Reviewer Identity:**
[REVIEWER_IDENTITY]

## INSTRUCTIONS

1. Evaluate the override justification against each criterion in the Minimum Sufficiency Criteria list.
2. For each criterion, determine whether it is SATISFIED, PARTIALLY SATISFIED, or NOT SATISFIED.
3. If any criterion is NOT SATISFIED, the overall assessment must be FAIL.
4. If all criteria are at least PARTIALLY SATISFIED but any remain PARTIALLY SATISFIED, the overall assessment is CONDITIONAL PASS.
5. Only return PASS if every criterion is SATISFIED.
6. For any criterion that is not fully satisfied, provide a specific, actionable description of what is missing or insufficient.
7. Do not evaluate whether the override decision itself was correct—only evaluate whether the justification is sufficiently documented.
8. If the justification references information not present in the provided context, flag this as a gap.

## OUTPUT SCHEMA

Return a JSON object with this exact structure:
{
  "overall_assessment": "PASS" | "CONDITIONAL PASS" | "FAIL",
  "criteria_evaluation": [
    {
      "criterion": "string (exact text from criteria list)",
      "status": "SATISFIED" | "PARTIALLY SATISFIED" | "NOT SATISFIED",
      "finding": "string (specific evidence or gap description)"
    }
  ],
  "missing_elements": ["string (list of specific missing information)"],
  "requires_reviewer_followup": true | false,
  "followup_questions": ["string (specific questions to ask the reviewer if followup is needed)"],
  "audit_readiness_note": "string (one-sentence summary of whether this justification would survive audit scrutiny)"
}

## CONSTRAINTS

- Do not invent or assume information not present in the justification text.
- Do not evaluate the correctness of the override decision.
- Flag vague language, unsupported claims, and missing rationale explicitly.
- For high-risk decisions, apply stricter scrutiny to completeness.
- If reviewer identity is missing or incomplete, flag this as a gap.

To adapt this template for your environment, start by defining your [SUFFICIENCY_CRITERIA] list. This should be a concrete, enumerated set of requirements such as: 'Override reason is stated in business terms,' 'Risk acknowledgment is present,' 'Alternative considered is documented,' and 'Reviewer identity and timestamp are recorded.' The criteria should be maintained as a configuration variable in your application, not hardcoded into the prompt, so that audit standards can evolve without prompt changes. The [RISK_LEVEL] input should map to your organization's risk taxonomy—for example, 'LOW,' 'MEDIUM,' 'HIGH,' or 'CRITICAL'—and the prompt will apply proportionally stricter scrutiny to higher-risk decisions.

When integrating this prompt into a production harness, validate the output JSON against the schema before routing. If the overall_assessment is FAIL or CONDITIONAL PASS, the harness should automatically route the justification back to the original reviewer with the followup_questions and missing_elements populated. Log every evaluation result alongside the justification ID for audit trail completeness. For high-risk or regulated domains, always include a human audit lead in the loop to review borderline CONDITIONAL PASS assessments before closing the override record. Never silently accept an insufficient justification—the entire purpose of this prompt is to prevent gaps from entering the permanent audit trail.

The override justification sufficiency evaluation prompt is designed to operate as a gate in an audit pipeline, not as a standalone chat interaction. When a human reviewer submits an override justification, the system should first run this prompt to determine if the justification meets minimum sufficiency criteria before accepting it into the permanent audit trail. The prompt produces a pass/fail assessment with specific gaps identified, which the harness uses to either accept the record or route it back to the reviewer with structured feedback on what's missing.

The implementation harness requires several components working together. First, define your minimum sufficiency criteria as a structured schema—typically requiring fields like override reason, risk acknowledgment, alternatives considered, and impact assessment. Pass this schema into the prompt's [SUFFICIENCY_CRITERIA] placeholder. Second, implement a validation layer that parses the model's JSON output and checks for the required fields: pass (boolean), gaps (array of strings), and recommendation (string). If the output fails schema validation, retry once with the validation error included in the prompt context. Third, build a routing decision: if pass is true, write the justification to the audit log with the evaluation metadata attached; if pass is false, return the gaps array to the reviewer's interface with a request for revision. Log every evaluation result—including retries and validation failures—for downstream audit trail completeness verification.

Model choice matters here. Use a model with strong instruction-following and structured output capabilities, such as GPT-4o or Claude 3.5 Sonnet, configured with a low temperature (0.0–0.2) to maximize consistency across evaluations. Do not use this prompt with small or locally fine-tuned models unless you've calibrated them against a golden dataset of pre-scored justifications. For high-stakes audit environments, implement a human review step for all borderline cases—specifically, when the model's confidence is implicit in vague gap descriptions or when the justification involves novel override categories not covered in your criteria. Store the raw model response alongside the parsed evaluation to support later pattern detection and bias analysis across override events. Finally, never allow an override to proceed without a passing sufficiency evaluation or explicit human auditor approval logged in the system.