Framework 08 of 12 · Post-execution Phase · Quality evaluation
Absent-State Audit
The discipline of judging what happened while you were away
A methodology for evaluating the quality of agentic experiences that unfold without human presence. Assesses integrity, outcome alignment, and return-state design. The discipline of judging what happened while you were away.
Absent-State Audit: Core Principles
Absence Is the Primary Use State
The most consequential agentic experiences happen when the human is not watching. This is not an edge case - it is the primary use state. The Absent-State Audit is designed for this reality, providing methodologies for evaluating experiences that unfolded without human presence.
Outcomes Must Be Evaluated Against Intent
The audit compares what happened against what was requested. Did the agent achieve the specified goals? Did it respect the specified constraints? Did it meet the success criteria? This intent-to-outcome alignment is the primary quality metric for absent-state experiences.
The Return State Must Be Designed
When the human returns from absence, they need to understand what happened, what changed, and what requires their attention. The return state is not a log dump - it is a designed experience that communicates the essential narrative of the agent's operation during absence.
Integrity Requires Independent Verification
The agent's self-report of its performance is necessary but not sufficient. The audit framework includes patterns for independent verification - checking outcomes against external sources, validating that constraints were respected, and confirming that the agent's account of its actions is accurate.
Audit Trails Must Be Complete and Immutable
Every action the agent takes during absence must be logged in an immutable audit trail. This trail serves multiple purposes: user review, regulatory compliance, system improvement, and trust calibration. Incomplete or mutable audit trails undermine every other aspect of the framework.
The most consequential experiences in agentic AI happen when no one is watching. The Absent-State Audit is the discipline of ensuring that what happened in your absence was what you would have wanted to happen in your presence.
Absent-State Audit: Implementation Patterns
Integrity Assessment
Methodologies for verifying that the agent acted within its delegated authority during absence. Includes scope compliance checks, constraint violation detection, and authority boundary verification against the original delegation.
When to use: After every absent-state operation, as part of the standard return-state experience.
Outcome Alignment Scoring
A scoring system that measures how well the agent's outcomes match the human's original intent. Includes goal achievement metrics, constraint adherence scores, and success criteria satisfaction rates. Produces a single alignment score that summarises the quality of the absent-state experience.
When to use: As a standard metric calculated for every completed delegation.
Return-State Design
Interface patterns for the moment when the human returns from absence. Includes executive summaries, decision highlights, exception reports, and action-required items. Designed to communicate the essential narrative in seconds, with progressive detail available on demand.
When to use: At every session start following an absent-state operation.
Absent-State Quality Metrics
A suite of metrics specifically designed for evaluating experiences that unfold without human presence. Includes time-to-completion, resource efficiency, exception handling quality, and deviation from plan. These metrics complement traditional UX metrics that assume human presence.
When to use: As ongoing measurement of agent performance during absent-state operations.
Audit Trail Architecture
Technical patterns for logging every agent action during absence in an immutable, queryable format. Includes action timestamps, decision rationale, context snapshots, and outcome records. Designed for both human review and automated compliance checking.
When to use: As a foundational infrastructure requirement for any agentic system.
Absent-State Audit: Commerce Applications
Transaction Review and Reconciliation
After the machine customer completes purchases during the consumer's absence, the Absent-State Audit provides a structured review experience: what was bought, from whom, at what price, how it compared to alternatives, and whether it met the specified criteria. This is not a transaction history - it is a quality assessment of the agent's purchasing decisions.
Delivery and Fulfilment Tracking
The audit extends beyond the purchase to fulfilment: were items delivered on time, in the expected condition, to the correct address? The framework provides patterns for tracking fulfilment quality as part of the overall absent-state assessment.
Budget Compliance Verification
For consumers who delegate with budget constraints, the audit verifies that spending remained within bounds. Includes per-transaction compliance, cumulative budget tracking, and alerts for approaching limits. Budget compliance is verified against the original delegation, not just the agent's self-report.
Alternative Analysis
The audit includes retrospective analysis of alternatives the agent considered but rejected. This transparency helps the human evaluate whether the agent's decision-making was sound - not just whether the outcome was acceptable, but whether better outcomes were available and why they were not chosen.
A transaction history tells you what the agent did. An Absent-State Audit tells you whether what the agent did was what you asked for, within the boundaries you set, achieving the outcomes you specified.
Absent-State Audit: Guidance for Teams
Start With
- -Define the return-state experience for your most common absent-state operation
- -Build an outcome alignment scoring system for your top 3 delegation types
- -Implement immutable audit trail logging for all agent actions
- -Create executive summary templates for absent-state reports
Build Toward
- -Automated anomaly detection that flags unusual absent-state behaviour
- -Comparative analysis showing how outcomes compare to historical performance
- -Predictive quality scoring that estimates outcome alignment before completion
- -Cross-agent audit coordination for multi-agent absent-state operations
Measure By
- -Outcome alignment score - how well do results match original intent?
- -Return-state comprehension time - how quickly do users understand what happened?
- -Audit trail completeness - are all agent actions logged without gaps?
- -Post-audit trust change - does the audit experience increase or decrease user trust?
Absent-State Audit: Lifecycle Connections
Framework 09
Explainability & Observability
The audit depends on explainability. Without transparent reasoning logs, the audit can only assess outcomes, not the quality of decision-making.
Explore frameworkFramework 04
Trust Calibration Model
The audit experience directly influences trust calibration. Good audits build trust; poor audits or hidden failures erode it.
Explore frameworkFramework 10
Failure Architecture Blueprint
When the audit reveals failures, the Failure Architecture Blueprint governs how those failures are communicated and resolved.
Explore frameworkAbsent-State Audit: What Comes Next
The Absent-State Audit evaluates outcomes. The next framework - Explainability & Observability - provides the transparency layer that makes meaningful auditing possible.
Absent-State Audit: The Framework Ecosystem
Navigate the complete lifecycle of Agentic Experience Design. Each framework addresses a distinct phase of the human-agent relationship.