burden
Burden Distribution Evals
Whether failure modes distribute burden equitably across user populations.
stable · 12 test cases· Est. 30 min
Open suiteGovernance eval suites
Test whether your AI system can be stopped, explained, appealed, and repaired — on a clock that matters.
Evals
Governance eval suites
Test whether your AI system can be stopped, explained, appealed, and repaired — on a clock that matters.
Available suites
Governance eval suites
Each suite tests a specific governability dimension. Run them individually or combine for a full assessment.
agency
Contestability Evals
Whether an LLM system's decisions can be effectively challenged and overturned.
stable · 10 test cases· Est. 25 min
Open suiteagency
Stoppability Evals
Whether humans can halt AI-driven processes at arbitrary points without catastrophic state loss.
stable · 10 test cases· Est. 20 min
Open suitetemporal
Temporal Rights Evals
Whether LLM systems respect the seven temporal rights from STD-01.
stable · 10 test cases· Est. 30 min
Open suitestructural
Reversibility Evals
Whether state changes made by LLM systems can be cleanly undone.
stable · 10 test cases· Est. 25 min
Open suitevisibility
Explainability-for-Accountability Evals
Whether LLM explanations are actionable for governance, not just decorative.
stable · 10 test cases· Est. 20 min
Open suitegovernance
Agent Governance Evals
Whether AI agents respect governance constraints during multi-step autonomous execution.
stable · 12 test cases· Est. 35 min
Open suiteburden
Cross-Domain Burden Index
Burden distribution across healthcare, finance, hiring, content moderation, and government services.
stable · 15 test cases· Est. 45 min
Open suiteMethodology
How evals are scored
Phase 1 scoring is manual and evidence-based. Each test case is evaluated by a human reviewer using a defined rubric.
Scoring anchors
- Binary: Pass/fail — condition is met or not.
- 0–3: Absent → Partial → Adequate → Strong.
- 0–5: Absent → Minimal → Basic → Adequate → Strong → Exemplary.
Scoring methods
- Weighted average: Each test case contributes according to its weight. Overall score must meet the passing threshold.
- Min-threshold: All test cases must score at or above the minimum threshold. Single critical failure fails the suite.
- All-must-pass: Every test case must pass. Used for high-stakes suites.
Manual scoring
In Phase 1, all evals are run manually. A human reviewer runs each test case, captures evidence, and assigns scores using the rubric. Automated runner support is planned for Phase 2.
Run an eval
Run a governance eval suite
Use the eval runner to step through test cases, capture evidence, and generate a scored report.
- Select a suite or run all suites for a full governability assessment.
- Each test case presents a prompt and system context for the reviewer.
- Score each test case using the provided rubric anchors.
- Attach evidence and notes for each test case.
- Generate a scored report with aggregate score, grade, and recommendations.