Web Frontend -- Phase 14 Addendum¶
This addendum extends the Web Frontend Guided Demo with Phase 14 evaluation and compliance features added to the React frontend.
New Dashboard Cards¶
Evaluation Metrics Card¶
The Dashboard page now includes an Evaluation Metrics stat card showing the latest CLEAR+ benchmark summary:
| Metric | Value | Description |
|---|---|---|
| Benchmarks | 3+ | Number of benchmark specs in the active suite |
| Pass Rate | 100% | Percentage of specs meeting all thresholds |
| Avg Efficacy | 0.95 | Mean efficacy across all benchmark specs |
| Avg Coverage | 0.90 | Mean persona coverage across all specs |
The card refreshes when new benchmark.completed events arrive via
the WebSocket connection.
Compliance Status Card¶
A Compliance Status card summarises the latest audit results:
| Metric | Value |
|---|---|
| Regulation | EU AI Act |
| Checks | 1,224 |
| Pass Rate | 96.4% |
| Warnings | 44 |
| High-Risk Personas | 8 |
Evaluation Page¶
The new Evaluation page (pages/Evaluation.tsx) provides an interactive
view of CLEAR+ benchmark results.
Radar Chart¶
A D3 radar chart (visualizations/RadarChart.tsx) renders all seven
CLEAR+ dimensions for a selected benchmark spec. Dimensions are arranged
radially with the value plotted as a filled polygon.
Benchmark History¶
A timeline chart shows CLEAR+ dimension trends across benchmark runs. Each run is a column; dimensions are colour-coded rows showing improvement or regression.
Threshold Overlay¶
The radar chart supports a threshold overlay that draws the pass/fail boundary as a dashed polygon. Dimensions falling below the threshold are highlighted in red.
Compliance Page¶
The new Compliance page (pages/Compliance.tsx) provides an interactive
view of audit results.
Risk Distribution¶
A donut chart shows the distribution of personas across EU AI Act risk tiers:
- HIGH (red) -- Personas with 3+ hard-stops or governance category
- LIMITED (amber) -- Decision-making roles or mandatory patterns
- MINIMAL (green) -- Default classification
Findings Table¶
A sortable table lists all audit findings with columns:
| Column | Description |
|---|---|
| Requirement | EU AI Act article reference |
| Status | PASS / FAIL / WARNING / N/A |
| Persona | Which persona was audited |
| Evidence | Evidence source and confidence |
| Remediation | Recommended fix (if any) |
Evidence Graph Visualisation¶
The evidence graph is rendered as a force-directed graph using the
existing ForceGraph component. Nodes are coloured by type:
- Requirements (blue)
- Evidence items (green)
- Personas (orange)
- Constitutions (purple)
WebSocket Events¶
Phase 14 adds new event types to the WebSocket feed:
| Event Type | Source | Payload |
|---|---|---|
benchmark.started |
BenchmarkRunner | Suite name, spec count |
benchmark.completed |
BenchmarkRunner | Results summary |
benchmark.regression |
BenchmarkRunner | Regression details |
compliance.audit.started |
CompliancePipeline | Regulation, persona count |
compliance.finding.raised |
CompliancePipeline | Requirement ID, status |
compliance.remediation.required |
CompliancePipeline | Action ID, priority |
compliance.audit.completed |
CompliancePipeline | Full audit summary |
The Protocol Explorer page now includes these events in its event feed.
Navigation Updates¶
The sidebar navigation has been updated with two new entries:
- Evaluation -- CLEAR+ benchmark results and radar charts
- Compliance -- Audit findings, risk distribution, evidence graphs
Tips¶
- Use the radar chart to quickly identify which CLEAR+ dimensions need attention for a given benchmark spec
- Filter the findings table by risk tier to focus on HIGH-risk personas
- Export evidence graphs via the download button for external tools
Related¶
- Web Frontend Guided Demo -- Base demo
- CLEAR+ Benchmark Demo -- CLI benchmark demo
- AI Act Audit Demo -- CLI audit demo