Explore · SVG gallery
SVG Template Gallery
Every chart template in the vfairness rendering module: 44 templates, shown as 49 example cards. Select any card for a full preview and the code that renders it.
How the 44 are counted
The machine count is stats.svg_templates in vfairness-manifest.json, which counts renderable charts only. The templates directory holds 46 .svg files (as of 2026-08-28): two are partials included by other templates, _shared_defs and _could_not_check, so list_templates() reports 46. The 49 cards show the reporting dashboard at its three tiers (executive, operational, technical) and three facade helpers (FairnessDashboard, MetricsStore, ReportGenerator). Every chart uses the Blanco design language: sharp corners, near-monochrome, with colour kept for four semantic tones (pass, warn, info, neutral).
SVG Rendering Module
The vfairness.rendering module provides polished, card-based SVG dashboards
using Jinja2 templates. SVG rendering needs the [rendering] extra (Jinja2): pip install vfairness[rendering]. The rendered SVG output is self-contained, with no runtime, browser, or JavaScript dependency.
Every chart on this page is synthetic demonstration data
Nothing in this gallery is a real evaluation of a real model. Each render is produced by
scripts/regenerate_gallery_examples.py, which drives the real library over invented
inputs so the pictures stay honest to the code, and none of it describes anybody's system.
That is a design property, not a disclaimer. A demonstration chart must never be mistakable for
an audit result, so the twelve renders that come from the library's own built-in demo fixtures
carry an EXAMPLE ONLY: SYNTHETIC DEMONSTRATION DATA, NOT A REAL EVALUATION
band on the canvas itself and repeat it at the front of the accessible <desc>,
so the marking survives being screenshotted, pasted into a slide, or read aloud by a screen
reader. Those fixtures are reachable only by passing example=True: an adapter called
with no data does not quietly fall back to the sample, it reports that it measured nothing.
Three states, never two
Every chart here distinguishes assessed-pass, assessed-fail and
could-not-check, and never collapses the third into either of the others. A row or
a chart that measured nothing gets no number, no badge, no colour and no plot point, and it is
counted in no total that would imply it was measured. Verdicts are graded over the graded subset
only, the count of ungraded items is stated on the canvas, and an unqualified all-clear is withheld
while anything is ungraded: a partly-run gate reads PASS with "2 passed, 0 failed,
1 not run" rather than ALL PASS. Several captions below note where a chart shows this
third state.
vfairness_0_svg_rendering_demo.ipynb Interactive walkthrough of every SVG template with live rendering and synthetic dataOpen it to see this run end to end. Every notebook is executed on each build, so the code here matches the library you installed.
0 of 0 example cards
Preprocessing & Feature Engineering
Bias detection and feature analysis visualizations
Bias Audit Dashboard
Comprehensive bias detection report with historical patterns, proxy variables, and risk scores.
bias_audit_to_svg(report)
Correlation Heatmap
Feature correlation matrix highlighting potential proxy relationships with protected attributes.
correlation_heatmap_to_svg(data)
Correlation Matrix
General-purpose feature correlation matrix with multi-method support (Pearson, Spearman, Cramér's V) and per-cell method indicators.
correlation_matrix_to_svg(data, methods=methods_dict)
Proxy Risk Analysis
Identifies features correlated with protected attributes that may enable indirect discrimination.
proxy_risk_to_svg(proxies)
Transformation Comparison
Compare fairness impact of different feature transformation methods.
transformation_comparison_to_svg(data)
Intersectional Analysis
Analyze fairness across intersectional subgroups to detect fairness gerrymandering.
intersectional_analysis_to_svg(data)
Intersectional Disparity
Ranked-bar chart with ground truth overlay, severity-colored bars, prediction deltas, and key insights. Dual-lens view comparing model predictions to ground truth across subgroups.
intersectional_disparity_to_svg(result)
Fairness Report Dashboard
High-level fairness summary with a score over the graded metrics and a pass/fail indicator per metric. The headline band also carries what was not graded: how many metrics could not be checked, and how many groups reported no rate and so are not plotted. A run that verified no metric against a threshold reads NOT ASSESSABLE and is given no score at all.
fairness_report_to_svg(report)
Data Validation
Pre-training data validation with severity-coded issues, quality metrics, and pass/fail gate status.
svg = data_validation_to_svg(data, explanation="...")
Auto-Discovery Scanner
Automated detection of protected attributes, fairness violations, and intersectional privilege analysis.
svg = auto_discovery_to_svg(data, explanation="...")
Training-Time Interventions
Fairness-aware training analysis and method comparison
Training Report
Compact dashboard with baseline metrics, method comparison, and recommendations.
training_report_to_svg(report)
Training Analysis Report
Full-page comprehensive analysis with group fairness bars, trade-off scatter, and action items.
training_analysis_report_to_svg(report)
Method Comparison
Bar chart comparing accuracy and fairness violation across training methods.
method_comparison_to_svg(comparisons)
Trade-off Analysis
Scatter plot showing accuracy vs fairness with Pareto frontier overlay.
tradeoff_analysis_to_svg(tradeoff_data)
Post-Processing Interventions
Threshold optimization and prediction reweighting
Threshold Optimization
Dashboard showing group-specific thresholds, positive rate distributions, and fairness improvement.
threshold_optimization_to_svg(result)
Reweighting Comparison
Method comparison showing fairness improvement, accuracy change, and calibration impact.
reweighting_comparison_to_svg(report)
CI/CD Pipeline Report
MLOps deployment gate report with fairness checks and approval status. The counts are taken over the checks that actually ran: a check whose status never arrived is named as not executed rather than folded into the executed total, and the all-clear wording is withheld while any check is ungraded, so a partly-run suite reads PASS with "2 passed, 0 failed, 1 not run" instead of ALL PASS.
cicd_pipeline_to_svg(result)
Calibration Visualizations
Group-specific calibration analysis and reliability diagrams
Calibration Report Dashboard
Comprehensive calibration analysis with ECE, MCE, and group-wise metrics.
calibration_report_to_svg(report)
Reliability Diagram
Classic reliability diagram showing predicted vs actual probabilities across bins, with ECE and MCE. Called with no data it reports that it could not check rather than reading out a flattering "Well Calibrated, ECE = 0.000" from an empty histogram.
reliability_diagram_to_svg(data)
Group Calibration
Overlaid calibration curves for each demographic group.
group_calibration_to_svg(data)
Calibration Disparity
Per-group calibration error with the spread between best and worst group. When no significance verdict is supplied the chart measures and prints the spread but withholds the verdict, stating that this is not a finding of no disparity; an overall ECE that was never computed reads N/A rather than 0.
calibration_disparity_to_svg(data)
Pareto Frontier
Accuracy-fairness-calibration trade-off frontier with optimal operating points.
pareto_frontier_to_svg(data)
Evaluation & Metrics
Fairness metrics visualization and statistical analysis
Fairness Detailed Report
Executive-level fairness assessment with score gauge, metrics table, and recommendations.
fairness_detailed_report_to_svg(report)
Radar Chart
Per-metric fairness on one axis: fair metrics reach outward into a round shape, failures cave toward the centre; a dashed ring marks the pass threshold.
radar_chart_to_svg(metrics)
Disparity Heatmap
Pairwise disparity matrix between all demographic groups.
disparity_heatmap_to_svg(disparities)
Metrics Bar Chart
Bar chart comparing multiple fairness metrics with threshold lines. The PASSED and FAIR RATE tiles count only the metrics that were actually checked against a threshold. A report with nothing gradeable in it is marked NOT ASSESSABLE and gets neither tile, rather than a reassuring 100%.
metrics_bar_chart_to_svg(metrics)
Group Comparison
Side-by-side comparison of metrics across demographic groups.
group_comparison_to_svg(data)
Effect Sizes
Cohen's d and other effect size measures with magnitude interpretations.
effect_sizes_to_svg(effects)
Confidence Intervals
Fairness metrics with the bootstrap or Bayesian confidence interval the report supplies. A metric that arrives without one is drawn as a bare point marked "CI not computed" and makes no significance claim, which is why the example above, whose report carries no intervals, shows that state on every row rather than a band. The chart never manufactures a band around a value.
confidence_intervals_to_svg(data)
Robustness Testing
Permutation tests, sensitivity analysis, and subgroup audit with overall robustness score and recommendations.
svg = robustness_testing_to_svg(data, explanation="...")
Ranking Fairness
Ranking fairness metrics with exposure parity, NDKL, group exposure bars, and position distribution table.
svg = ranking_fairness_to_svg(data, explanation="...")
Regression Fairness
Regression equity report with MAE/RMSE/R² parity badges, per-group metrics, effect sizes, and residual bias.
svg = regression_fairness_to_svg(data, explanation="...")
Monitoring & Operations
Real-time fairness monitoring, drift detection, and alert visualizations
Monitoring Dashboard
Live fairness monitoring snapshot with metric values, alert status, group rates, and distribution shift (MMD) scores.
monitoring_dashboard_to_svg(window_metrics)
Drift Report
Multi-scale fairness drift analysis with KS statistics, p-values, mean shifts, and per-scale detection status.
drift_report_to_svg(drift_result)
Alert Timeline
Chronological timeline of monitoring events showing alert status, sample counts, and flagged metrics.
alert_timeline_to_svg(history)
Temporal Analysis
Trend analysis with metric summaries, weekly patterns, slope detection, and degradation checks over time.
temporal_analysis_to_svg(analyzer)
Reporting & Dashboards
Interactive dashboards, automated reports, and metrics storage for stakeholder communication
Executive Dashboard
Board-level fairness summary: green health score (91/100), all metrics passing, stable trend. Minimal detail for executive stakeholders.
svg = reporting_dashboard_to_svg(tier="executive", example=True, save_path="exec.svg")
Operational Dashboard
Engineer-level dashboard: yellow health score (72/100), 1 threshold breach, 3 active alerts, degrading trend. Full breach report and NLG narrative.
svg = reporting_dashboard_to_svg(tier="operational", example=True, save_path="ops.svg")
Technical Audit
Full statistical audit: red health score (38/100), 3 metrics breached, 7 alerts (3 critical), confirmed drift event. KS stats and projected time-to-breach.
svg = reporting_dashboard_to_svg(tier="technical", example=True, save_path="tech.svg")
Metrics Store Dashboard
Unified data layer with privacy-preserving queries, health scores, and metric ingestion from monitors and experiments. The preview image is illustrative; MetricsStore is a data layer, not a chart.
store = MetricsStore(); store.ingest_from_monitor(monitor)
Interactive Fairness Dashboard
Plotly-based progressive-disclosure dashboard with executive, engineer, and auditor views. Supports time-window filtering. The preview image is illustrative; the class produces interactive Plotly output, not this SVG.
dashboard = FairnessDashboard(store); fig = dashboard.create_executive_view()
Automated Report Generator
Multi-format, multi-tier report generation with natural language summaries. Supports executive, engineer, and auditor tiers in HTML, PDF, and JSON. The preview image is illustrative; the class emits HTML, PDF, and JSON reports, not this SVG.
gen = ReportGenerator(store, dashboard); report = gen.generate_executive_report()
Experimentation & A/B Testing
Fairness-aware A/B testing, power analysis, Pareto optimization, and causal decomposition
Experiment Results
A/B test results with overall effect, forest plot of per-intersection effects, confidence intervals, and heterogeneity test.
svg = experiment_results_to_svg(result_data, explanation="...")
Experiment Recommendation
Deployment decision dashboard with confidence gauge, reasoning list, trade-off table, and caveats panel.
svg = experiment_recommendation_to_svg(rec_data, explanation="...")
Power Analysis
Per-intersection statistical power analysis with bar chart, 80% threshold, and minimum detectable effect summary.
svg = power_analysis_to_svg(power_data, explanation="...")
Causal Decomposition
Causal mediation analysis with direct/indirect effects, Baron-Kenny steps, proportion mediated gauge, and temporal stability.
svg = causal_decomposition_to_svg(decomposition, temporal, explanation="...")
Workflow Integration
MLOps, CI/CD gates, and PR report cards
Workflow Overview
Development workflow integration pipeline showing experiment tracking, version control, CI/CD gates, testing, and collaborative review stages. The render above is the built-in reference set, reached with
example=True and watermarked as such; called with no pipeline described, the chart reports that nothing was assessed instead of presenting the sample as your workflow.svg = workflow_overview_to_svg(example=True, explanation="...")
Hierarchical Fairness Gate
Three-level hierarchical fairness gate showing overall, single-attribute, and intersectional evaluation with small-sample warnings. Each level is graded over the metrics that were measured at that level, and a call with no decision renders NOT CHECKED rather than an approval.
svg = hierarchical_gate_to_svg(decision, explanation="...")
Fairness Report Card
PR-ready fairness report card with metric results, blocking issues, warnings, and deployment recommendation. The verdict comes from the gate decision you pass in:
report_card_to_svg(None) renders NOT CHECKED, and cannot reach the built-in sample or print an approval from no data.svg = report_card_to_svg(decision=gate_decision, model_name="model-v3")
Complete Template Reference
Every SVG template in vfairness.rendering plus interactive components from vfairness.operations.reporting and vfairness.operations.experimentation:
| Template | Adapter Function | Category | Description |
|---|---|---|---|
bias_audit.svg | bias_audit_to_svg() | Preprocessing | Bias detection dashboard |
correlation_heatmap.svg | correlation_heatmap_to_svg() | Preprocessing | Feature correlation matrix |
correlation_matrix.svg | correlation_matrix_to_svg() | Preprocessing | General multi-method correlation matrix |
proxy_risk.svg | proxy_risk_to_svg() | Preprocessing | Proxy variable detection |
transformation_comparison.svg | transformation_comparison_to_svg() | Preprocessing | Feature transformation impact |
intersectional_analysis.svg | intersectional_analysis_to_svg() | Preprocessing | Subgroup analysis |
intersectional_disparity.svg | intersectional_disparity_to_svg() | Preprocessing | Ranked intersectional disparity, ground truth vs prediction |
fairness_report.svg | fairness_report_to_svg() | Preprocessing | Summary dashboard |
data_validation.svg | data_validation_to_svg() | Preprocessing | Pre-training data validation |
auto_discovery.svg | auto_discovery_to_svg() | Preprocessing | Protected attribute scanner |
training_report.svg | training_report_to_svg() | Training | Compact training report |
training_analysis_report.svg | training_analysis_report_to_svg() | Training | Full training analysis |
method_comparison.svg | method_comparison_to_svg() | Training | Method bar chart |
tradeoff_analysis.svg | tradeoff_analysis_to_svg() | Training | Pareto scatter plot |
threshold_optimization_report.svg | threshold_optimization_to_svg() | Post-Processing | Threshold dashboard |
reweighting_comparison_report.svg | reweighting_comparison_to_svg() | Post-Processing | Reweighting methods |
cicd_pipeline.svg | cicd_pipeline_to_svg() | Post-Processing | Deployment gate |
calibration_report.svg | calibration_report_to_svg() | Calibration | Calibration dashboard |
reliability_diagram.svg | reliability_diagram_to_svg() | Calibration | Classic reliability plot |
group_calibration.svg | group_calibration_to_svg() | Calibration | Per-group calibration |
calibration_disparity.svg | calibration_disparity_to_svg() | Calibration | Calibration gap |
pareto_frontier.svg | pareto_frontier_to_svg() | Calibration | Trade-off frontier |
fairness_detailed_report.svg | fairness_detailed_report_to_svg() | Evaluation | Executive report |
radar_chart.svg | radar_chart_to_svg() | Evaluation | Multi-metric radar |
disparity_heatmap.svg | disparity_heatmap_to_svg() | Evaluation | Pairwise disparities |
metrics_bar_chart.svg | metrics_bar_chart_to_svg() | Evaluation | Metrics comparison |
group_comparison.svg | group_comparison_to_svg() | Evaluation | Group metrics |
effect_sizes.svg | effect_sizes_to_svg() | Evaluation | Effect magnitudes |
confidence_intervals.svg | confidence_intervals_to_svg() | Evaluation | Statistical CIs |
robustness_testing.svg | robustness_testing_to_svg() | Evaluation | Robustness & sensitivity testing |
ranking_fairness.svg | ranking_fairness_to_svg() | Evaluation | Ranking fairness metrics |
regression_fairness.svg | regression_fairness_to_svg() | Evaluation | Regression equity report |
monitoring_dashboard.svg | monitoring_dashboard_to_svg() | Monitoring | Live fairness dashboard |
drift_report.svg | drift_report_to_svg() | Monitoring | Multi-scale drift report |
alert_timeline.svg | alert_timeline_to_svg() | Monitoring | Alert history timeline |
temporal_analysis.svg | temporal_analysis_to_svg() | Monitoring | Trend & pattern analysis |
experiment_results.svg | experiment_results_to_svg() | Experimentation | A/B test results & forest plot |
experiment_recommendation.svg | experiment_recommendation_to_svg() | Experimentation | Deployment recommendation |
power_analysis.svg | power_analysis_to_svg() | Experimentation | Statistical power analysis |
causal_decomposition.svg | causal_decomposition_to_svg() | Experimentation | Causal mediation analysis |
workflow_overview.svg | workflow_overview_to_svg() | Workflow | Development workflow pipeline |
hierarchical_gate.svg | hierarchical_gate_to_svg() | Workflow | Hierarchical fairness gate |
report_card.svg | report_card_to_svg() | Workflow | PR fairness report card |
reporting_dashboard.svg | reporting_dashboard_to_svg(tier="executive"|"operational"|"technical") | Reporting | Multi-tier fairness report dashboard |
MetricsStore | MetricsStore() | Reporting | Unified metrics data layer |
FairnessDashboard | FairnessDashboard() | Reporting | Interactive Plotly dashboard |
ReportGenerator | ReportGenerator() | Reporting | Automated multi-format reports |
FairnessExperiment | FairnessExperiment() | Experimentation | Fairness A/B testing |
FairnessPowerAnalyzer | FairnessPowerAnalyzer() | Experimentation | Power analysis & SPRT |
ExperimentAnalysis | ExperimentAnalysis() | Experimentation | Pareto & causal analysis |
Quick Usage
python
from vfairness.rendering import (
render_svg, list_templates,
# Report dashboards
bias_audit_to_svg, calibration_report_to_svg, fairness_report_to_svg,
# Feature engineering
correlation_matrix_to_svg,
# Training
training_report_to_svg, training_analysis_report_to_svg,
# Post-processing
threshold_optimization_to_svg, reweighting_comparison_to_svg,
# Evaluation
fairness_detailed_report_to_svg, radar_chart_to_svg,
# Monitoring
monitoring_dashboard_to_svg, drift_report_to_svg,
alert_timeline_to_svg, temporal_analysis_to_svg,
# Reporting
reporting_dashboard_to_svg,
# Experimentation
experiment_results_to_svg, experiment_recommendation_to_svg,
power_analysis_to_svg,
# Workflow integration
workflow_overview_to_svg,
hierarchical_gate_to_svg,
report_card_to_svg,
)
# List all available templates
print(list_templates())
# Render using adapter functions
svg = training_analysis_report_to_svg(report, save_path='report.svg')
# Or use render_svg directly
svg = render_svg('radar_chart', {'metrics': [...], 'groups': [...]})
# ── Reporting ──────────────────────────────────────────────
from vfairness.operations.reporting import (
MetricsStore, FairnessDashboard, ReportGenerator, InteractiveDashboard
)
store = MetricsStore()
store.ingest_from_monitor(monitor) # Feed from monitoring
dashboard = FairnessDashboard(store)
fig = dashboard.create_executive_view() # Plotly figure
gen = ReportGenerator(store, dashboard)
report = gen.generate_executive_report() # HTML / PDF / JSON
# ── Experimentation ────────────────────────────────────────
from vfairness.operations.experimentation import (
FairnessExperiment, FairnessPowerAnalyzer, ExperimentAnalysis
)
exp = FairnessExperiment(
control_data=df_ctrl, treatment_data=df_treat,
protected_attributes=['gender', 'race'], outcome_column='approved',
)
result = exp.run_full_analysis()
analysis = ExperimentAnalysis(result, experiment=exp)
rec = analysis.decision_recommendation() # Deploy / hold / revert