Frontend dashboard visualization

Start with the user's analytical question and the real data shape, then choose the simplest encoding that supports an accurate decision.

When to invoke

• "Design an operational dashboard from these real metrics."
• "Choose the right chart for this analytical question."
• "Review this KPI page for misleading or inaccessible visualization."
• "Define dashboard filters, tables, drilldowns, and failure states."
• "Test a chart-heavy frontend with realistic edge cases."

Analytical contract

For every metric or visualization, record:

• question or decision supported;
• metric name, definition, unit, time window, comparison baseline, update time, and provenance;
• dimensions, cardinality, ordering, missingness, precision, and uncertainty;
• target audience and required scanning or exact-value behavior;
• filters, drilldowns, exports, annotations, thresholds, and access limits;
• loading, empty, delayed, partial, stale, unavailable, error, and recovery states.

Never invent a metric or use fictional production values unless the user explicitly requests labeled synthetic fixture data.

Chart and table selection

Read references/chart-selection.md. Prefer tables for exact values and operational scanning. Use charts only when position, length, area, color, or another encoding improves the stated question.

Read references/dashboard-patterns.md for filter, provenance, density, and state behavior. Read references/accessible-data-visualization.md for summaries, tables, keyboard, focus, and non-color alternatives.

Criteria

• Titles and summaries reflect active filters, time windows, and comparison context.
• Zero, missing, delayed, partial, stale, and unavailable data remain distinct.
• Dates, timezones, currency, percentages, precision, and large values are consistent.
• Dense labels, negative values, outliers, long series, empty ranges, and loading do not create misleading output.
• Interactive controls have accessible names, keyboard behavior, visible state, and reversible outcomes.
• Color is not the only series, threshold, selection, or status cue.
• Download or underlying tables appear only when product and data policy permit them.
• Rendering cost, large collections, and update frequency fit the product's runtime budget.

Use assets/operational-dashboard-review.md and assets/human-review-checklist.md.

Limits

• Do not select a chart because it is fashionable or available in a library.
• Do not infer causation from correlation or precision from decorative animation.
• Do not add a new chart library when the installed one satisfies the contract.
• Do not expose restricted raw data or enable exports without product policy.

Output template

## Dashboard visualization result
**Status:** ready | needs revision | blocked

### Analytical contract
| Question | Metric/data | Context and provenance | Audience action |
| --- | --- | --- | --- |

### Encodings and interactions
| Need | Selected table/chart | Why | Accessible alternative | Edge states |
| --- | --- | --- | --- | --- |

### Validation
| Fixture/state | Behavioral check | Visual/a11y check | Result |
| --- | --- | --- | --- |

Quality gate

• Every visualization traces to a real analytical question and data shape.
• Metrics include units, windows, baselines, freshness, and provenance.
• Zero, missing, partial, delayed, stale, unavailable, and error states remain distinguishable.
• Exact-value, accessible-summary, keyboard, color-independent, and edge-data behavior is defined.
• Library and rendering-cost decisions use repository evidence.
• The human review checklist has no unresolved blocked item.