aditya-ariosity Dashboard Redesign
Design a decision instrument, not a wall of metrics. Preserve useful density, expose the comparisons that matter, and make data quality and scope legible.
1. Establish the decision model
Identify:
- users: primary role, expertise, frequency, environment, devices, display duration, accessibility, and export needs;
- decisions: required actions, monitoring cadence, acceptable time to notice, and the workflow from signal through investigation to action;
- data: metric definitions, owners, sources, freshness, and latency;
- context: required comparisons, targets, thresholds, and segments.
If these are unknown, state assumptions and mark metric semantics for owner validation. Never infer that a visible number is a KPI.
Checkpoint: State assumptions, have the data owner validate metric meaning and thresholds, revise the decision matrix and affected representations, then re-check until every unresolved semantic question is explicitly marked for decision.
Classify the surface using references/dashboard-models.md. A monitoring board, executive scorecard, operational queue, and analytical workspace need different density and interaction.
2. Inventory decisions before widgets
Build a compact matrix:
Decision | Signal | Comparison | Grain | Freshness | Action | Failure cost
Synthetic example: Which route needs intervention? | Risk state | Operations threshold and shift target | Route | Per vehicle | Open incident | Missed service
Remove or demote data that does not support a scoped decision, diagnosis, or required reporting obligation. Identify missing context before selecting chart types.
Checkpoint: Every row ends in a real decision or reporting obligation; remove the rest.
3. Create the information architecture
Organize the page in this order when it matches the task:
- scope and data state;
- conditions requiring attention;
- core outcome and comparison;
- causes or contributing dimensions;
- records, evidence, and actions.
Use stable regions and consistent reading order. Keep controls near the data they affect. Make global and local filters visually distinct. Preserve active filters, time zone, time range, currency, units, and comparison basis across drilldown.
Do not turn every measure into a card. A compact table or sentence can outperform a chart; a chart can outperform a single number when pattern or comparison matters.
Checkpoint: Trace each region to a matrix row; revise orphaned or duplicated regions.
4. Select representations
Read references/visualization-and-metrics.md. For each element, state:
- the question it answers;
- why this representation is faster or safer than the alternatives;
- the comparison and baseline;
- encoding, units, precision, labels, and ordering;
- accessible equivalent or data access;
- behavior for missing, partial, stale, or delayed data.
Prefer position and length for precise comparison. Avoid 3D, decorative gauges, unbounded color ramps, and charts whose meaning depends on hover. Use direct labels when practical; do not rely on color alone.
Checkpoint: Verify each element's question, comparison, data state, and accessible equivalent; replace failures.
5. Design interactions and drilldown
Define the path:
overview -> notice -> compare -> isolate -> inspect -> act -> confirm -> return
Specify filtering, cross-filtering, sorting, pagination or virtualization, selection, zoom, annotations, saved views, sharing, export, and reset only when the user needs them. Keep the transition from aggregate to record-level evidence reversible and context-preserving.
For operational dashboards, expose ownership, status, age, urgency, and next action. For analytical workspaces, support comparison and hypothesis testing without hiding definitions or silently changing denominators.
Checkpoint: Walk the drilldown forward and back; fix any lost scope, state, or recovery.
6. Cover states and constraints
Read references/interaction-accessibility-qa.md. Define:
- loading, streaming, stale, partial, empty, no-result, error, permission, and offline states;
- responsive priorities and table/chart adaptation;
- keyboard and screen-reader behavior;
- reduced motion and non-color signals;
- latency feedback and expensive-query behavior;
- data lineage, freshness, and definition access.
Checkpoint: Verify every required state has visible and assistive behavior; add gaps before writing the brief.
7. Produce the redesign brief
For a complete input-to-output demonstration, read references/worked-example.md. Use it to calibrate decision architecture and specification depth; do not reuse its fictional metrics or thresholds.
Return:
Diagnosis
Name the dashboard type, primary decisions, dominant failure modes, and evidence confidence in no more than five bullets.
Decision architecture
Provide the decision matrix and proposed section order.
Component specification
For each region use:
Purpose | Content | Representation | Interaction | States | Responsive rule | Accessibility
Synthetic example: Prioritize interventions | Route, risk, age, freshness, owner | Sortable table | Select route; open incident | Empty, stale, permission | Preserve identity, risk, and action | Named sortable headers; non-color status
Metric and chart rationale
Explain additions, removals, merges, and representation changes. Separate definition questions from visual design.
Drilldown contract
Describe scope persistence, transition, detail, actions, and return behavior.
Acceptance criteria
Write observable criteria for comprehension, data correctness, interaction, accessibility, responsiveness, and performance.
Validation plan
Recommend task-based testing with representative users and real or realistic data. Include success signals such as time to detect, interpretation accuracy, decision confidence, error rate, and action completion. Do not use generic satisfaction as the sole measure.
Quality bar
- every visible element supports a decision, diagnosis, action, or obligation;
- metrics expose definition, units, scope, comparison, and freshness where material;
- hierarchy follows urgency and task, not visual novelty;
- density reflects expertise and frequency;
- chart choice matches the analytical question;
- filters and drilldowns preserve context;
- missing and stale data cannot masquerade as zero or current;
- unsupplied metrics, thresholds, baselines, viewport values, density, or styling remain unresolved inputs or labeled proposals, never facts;
- accessibility does not depend on a separate simplified dashboard.