adr-deterministic-mock-evaluation-harness
Decision
Promote deterministic-mock-evaluation-harness to an authoritative global architecture pattern following Reference Architecture ratification.
Context
- Source: governance-flow
- Pattern: deterministic-mock-evaluation-harness
- Promotion Recommendation: recommendation-deterministic-mock-evaluation-harness
- Scope Classification: global
- Scope Classification Artifact: pattern-scope-classification
The pattern encapsulates drop-in mock LLM capabilities allowing deterministic, zero-cost, and hermetic execution of agentic workflows and governance decision flows within CI/CD pipelines without external API flakiness.
Scope
Applicability: Applies across all agent orchestration pipelines, offline evaluation suites, and continuous architecture verification gates.
Global application to continuous testing and governance validation environments.
Consequences
Governance
Formalises offline deterministic verification as a required governance gate for continuous architecture pipelines.
Execution
Eliminates network latency, provider flakiness, and operational API costs during routine test and CI runs.
System Behaviour
Ensures agent evaluation runs predictably and idempotently under reproducible mock conditions.
Evidence Summary
Promotion Evidence
Ratified Reference Architecture in ArchGov and verified implementation in src/cas_runner/llm/mock_client.py.
Promotion Criteria
Satisfies governance maturity and ratified reference architecture requirements.
Governance Risks
No critical governance risks identified.
Traceability
Pattern Proposal
deterministic-mock-evaluation-harness
Scope Classification
Artifact: pattern-scope-classification
Classification: global
Promotion Recommendation
recommendation-deterministic-mock-evaluation-harness
Promotion Decision
promotion-decision-deterministic-mock-evaluation-harness
Next Step
- Submit ADR to Architecture Forum for formal institutionalisation
- Generate global pattern record in knowledge/global/patterns/