01The Foundational Data Structure
A Trusted Work Unit is a cryptographically sealed record of completed work that captures inputs, processing, outputs, and attribution in a standardized, immutable format. TWUs create an audit trail that can be verified independently and used for compliance, billing, and accountability purposes.
The Trusted Work Unit is not merely another log format. It represents a paradigm shift in how we think about work records. Traditional logs capture events. TWUs capture meaning—the complete context of what happened, why, and who or what was responsible.
This article examines the TWU architecture in detail, explaining each component and how they combine to create trustworthy work records.
02Core Components
The Work Identifier establishes uniqueness and enables correlation. Each TWU receives a unique identifier combining a timestamp, source system identifier, and random component. This format ensures global uniqueness while enabling temporal ordering and source identification. The identifier becomes the primary key for all references to this work unit.
Temporal anchoring captures when work occurred with precision sufficient for regulatory requirements and forensic analysis. The timestamp uses ISO 8601 format with timezone and microsecond precision. The window indicates whether this represents an instantaneous event or bounded time period.
03Attribution Architecture
Attribution represents perhaps the most innovative aspect of TWUs. Rather than treating work as either human or AI, TWUs capture the precise contribution of each participant.
Participants are listed with their type (human, AI model, or hybrid system), identifier, role in the work, and contribution percentage. The total must sum to one hundred percent.
Human attribution captures individuals who contributed. The identifier links to the identity management system. The role describes the function—reviewer, decision-maker, input provider. The contribution percentage represents the estimated share of the outcome attributable to this human.
AI attribution captures AI systems that contributed. The model field identifies the specific model version. The role describes the function—recommender, generator, analyzer. The contribution captures the share attributable to this AI. Provider identifies the AI service provider if external.
04The Evidence Chain
The evidence chain distinguishes TWUs from simple logs. Rather than recording only the final outcome, TWUs capture the complete sequence of events that produced the outcome.
Each evidence step has a sequence number, timestamp, actor, action type, description, and optional supporting data. The step sequence provides ordering. Actor identifies who or what performed this step. Action type categorizes the step—input, processing, validation, decision. Description provides human-readable explanation. Data contains structured information supporting the step.
05Quality Metrics
TWUs incorporate quality assessment enabling filtering, analysis, and compliance verification. Metrics include a score from zero to one hundred, confidence level, and dimensional breakdown.
Dimensions break quality into components: accuracy, completeness, timeliness, and compliance. This multidimensional view enables sophisticated quality analysis.
06Input and Output Capture
Complete input and output capture enables reconstruction and verification.
Inputs include primary and secondary inputs with type, source, timestamp, and content. The content hash enables verification without storing raw content when privacy requires.
Outputs include the primary output with type, content, and hash, plus outcome description, outcome type classification, and business outcome information like revenue impact.
07Cryptographic Integrity
The integrity structure provides tamper evidence and verification capability. The sealed hash covers the entire TWU content using SHA-256, created at the specified timestamp using a versioned algorithm.
Verification enables independent confirmation that the TWU has not been modified since sealing.
08Regulatory Alignment
TWUs can reference regulatory frameworks and compliance status. Framework identifies the applicable regulation. Requirement specifies the relevant requirement. Status indicates whether fulfilled, partial, or pending. Evidence reference points to supporting documentation.
09Implementation Patterns
Synchronous capture integrates TWU creation into the work process itself. When work completes, the TWU is created as part of the transaction. This ensures completeness but adds latency to the work process.
Asynchronous capture separates TWU creation from work execution. Events are captured during work, and TWU assembly happens afterward. This minimizes latency impact but requires careful event correlation.
Streaming capture handles high-volume scenarios. Events flow to a stream processor that assembles TWUs in near-real-time. This scales efficiently but requires infrastructure for stream processing.
10Platform Implementation
AI governance platforms like Veratrace provide TWU infrastructure including schema enforcement ensuring TWU validity, integrity services providing cryptographic sealing, storage optimized for TWU access patterns, query capabilities for TWU retrieval and analysis, and compliance reporting built on TWU data.
The goal is making TWU creation seamless for integrated AI systems while providing robust infrastructure for the resulting records.
11Conclusion
Trusted Work Units represent a new standard for work records. By combining comprehensive attribution, evidence chains, quality metrics, and cryptographic integrity, TWUs enable governance, compliance, and accountability for AI-enabled work.
Organizations building AI systems should consider TWUs as foundational infrastructure rather than an afterthought. The architecture supports current compliance needs while preparing for regulatory requirements that will only become more demanding.
