borghei/Claude-Skills / project-management/discovery/metrics-dashboard
metrics-dashboard
VerifiedDesigns a product metrics dashboard in four layers (North Star, inputs, guardrails, operational) with owners, comparisons and cadence, and audits the spec. Use when building a launch dashboard or pruning one nobody uses.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.1 KB |
| assets/dashboard_spec_template.md | markdown | 1.6 KB |
| assets/sample_dashboard_spec.json | config | 2.0 KB |
| assets/sample_dashboard_spec_flawed.json | config | 447 B |
| references/dashboard-anti-patterns.md | markdown | 4.6 KB |
| references/dashboard-architecture.md | markdown | 7.9 KB |
| scripts/dashboard_designer.py | script | 11.0 KB |
references/dashboard-architecture.md
View on GitHub# Dashboard Architecture Reference
## 1. The dashboard layers
```
┌──────────────────────────────────────────┐
│ NORTH STAR (1) │
│ ──────────────── │
│ [headline metric + trend + vs target] │
└──────────────────────────────────────────┘
┌──────────────────────────────────────────┐
│ INPUTS (3-5) │
│ ──────────── │
│ [3-5 chart cards, each driving NS] │
└──────────────────────────────────────────┘
┌──────────────────────────────────────────┐
│ GUARDRAILS (3-5) │
│ ────────────── │
│ [counter-metrics; "what we don't want │
│ to sacrifice"] │
└──────────────────────────────────────────┘
┌──────────────────────────────────────────┐
│ OPERATIONAL (4-8 per team) │
│ ─────────── │
│ [team-specific metrics for weekly │
│ action] │
└──────────────────────────────────────────┘
```
## 2. Per-layer design
### North Star layer
- Single metric, displayed prominently
- Comparison: vs prior period + vs target
- Trendline showing direction
- Annotated with major events (releases, campaigns)
### Inputs layer
- 3-5 metrics laid out as cards
- Each labeled "drives [NS]"
- Same comparisons
- Smaller charts (don't compete with NS visually)
### Guardrails layer
- 3-5 counter-metrics
- Distinct color (red / orange) to signal "watch out"
- Threshold alerts when crossed
- Often less-frequent refresh
### Operational layer
- 4-8 per team
- Live or daily refresh
- Team-specific framing
- Filterable by segment / channel / cohort
## 3. Audience-specific dashboards
### Exec dashboard (board / monthly business review)
- NS layer + inputs + 1-2 strategic guardrails
- Trends: month-over-month, quarter-over-quarter
- Highly summarized
- 1 page max
### Functional team dashboard (eng / growth / CS)
- NS (reminder) + their inputs + their operational metrics
- Daily / hourly refresh
- Drill-down capability
- 1-2 pages
### Individual contributor view
- Their KPIs (1-3)
- Their pipeline / queue / activity
- Personal — not exposed across team
### All-hands / company-wide
- NS + 3-5 inputs
- Mostly narrative (story per metric)
- 1 page; high signal/noise
## 4. Cadence by metric type
| Metric type | Cadence |
|-------------|---------|
| Real-time operational (active users now, errors) | Live |
| Daily ops (deploys, support tickets) | Daily |
| Weekly review (funnel, cohort, growth) | Daily refresh; weekly review |
| Monthly business review | Monthly refresh; monthly review |
| Quarterly board / strategy | Monthly refresh; quarterly review |
Don't refresh real-time what doesn't need it — it's expensive and noisy.
## 5. Common visualization patterns
### Time series (line)
- For: NS, inputs, most operational
- Comparison: same metric, prior period overlaid
- Annotations: deployment markers, campaign launches
### Funnel
- For: conversion / activation / checkout
- Comparison: vs cohort baseline / vs target
- Drill-down: by segment, channel
### Cohort heatmap
- For: retention, engagement over time
- Comparison: across cohorts (down a column)
- Read: vertically to spot improvement
### Bar / grouped bar
- For: comparison across discrete groups
- Comparison: same group prior period
### Bullet chart
- For: actual vs target gauge
- One per goal
### Sparkline (small embedded line)
- For: many metrics at-a-glance
- Comparison: implied trend
### What to avoid
- Pie charts beyond 3 slices
- 3D anything
- Default Excel charts (low information density)
- Dual y-axis (confusing)
## 6. Metric naming
Naming matters more than people think:
### Good
- "Activation Rate (W1)" — clear, specific, time-bounded
- "Daily Active Companies / MAC" — specific cohort, definition implied
- "P95 Lead Time" — specific statistic
### Weak
- "Engagement" — what kind?
- "Health Score" — composite; opaque
- "User Activity" — too vague
- "Volume" — of what?
Per metric, document:
- Name (precise)
- Definition (the exact calculation)
- Source (which table / event)
- Filters applied
- Caveats (known biases)
## 7. Metric ownership
Each metric needs a named owner:
- Owner is accountable for the metric moving (or for understanding why it's not)
- Owner answers questions about definition / data quality
- Owner authorizes changes to the metric
Without owners, metrics rot.
## 8. Threshold alerts
For guardrails and critical operational metrics:
- Define normal range
- Alert when outside range (3-sigma; or absolute threshold)
- Route to owner
- Acknowledge required
Don't over-alert. Alert fatigue kills response discipline.
## 9. Drill-down design
A good dashboard supports:
- **Click metric → see definition + history**
- **Filter by segment / channel / cohort**
- **Time-range adjuster**
- **Export to CSV / share link**
- **Annotations layer**
Without drill-down, the dashboard answers some questions and creates more.
## 10. Anti-vanity test
For each metric on the dashboard, ask:
1. "If it moved +10% next week, what would we do?"
2. "If it moved -10% next week, what would we do?"
If both answers are "nothing" → vanity → cut.
## 11. Tooling
Common dashboard tools (2026):
- Looker (Google) — strong semantic layer
- Mode — analyst-focused
- Hex — notebook + dashboard hybrid
- Sigma — spreadsheet-flavored
- Metabase — open-source
- Superset — open-source
- Tableau — enterprise classic
- PowerBI — enterprise + Microsoft shop
- Custom (Grafana + Metabase + Looker mix)
Tool < discipline. A spreadsheet dashboard reviewed weekly > a Looker
masterpiece nobody opens.
## 12. The 4-week pilot
Before commiting a new dashboard:
- Week 1-2: build with team input
- Week 3: weekly review attended
- Week 4: cut metrics nobody asked about
- Week 5+: locked in
Most dashboards bloat over time. Annual pruning is healthy.
## 13. Common pitfalls
- **30+ metrics, no hierarchy.** No prioritization possible.
- **No NS.** Optimizing nothing.
- **No guardrails.** Optimizing NS at any cost.
- **Live everything.** Cost + noise.
- **No owners.** Metric rot.
- **No comparisons.** Numbers in space.
- **Pretty charts, no decisions.** Decoration.
- **Built once, never reviewed.** Dashboard rot.
---
## 14. Audit limits by audience and the vanity test
These are the limits `dashboard_designer.py` applies.
### Top-level metric count (NS + inputs + guardrails)
| `audience` | Max top-level | Why |
|------------|---------------|-----|
| `exec` (alias `board`) | 8 | Limited attention; high signal/noise |
| `team` (alias `functional`) | 12, plus up to 8 operational per team | Acted on weekly |
| `all-hands` | 5 | Communicable; the company rallies on it |
| `partner` | 6 | External; share only what you will defend |
| `ic` | 3 | Their direct impact |
### Vanity vs actionable test
For each candidate metric: "If this moved 10% next week, what would we do?" No answer means it is vanity: cut it. Every chart also needs a comparison anchor (prior period, target or cohort baseline). Visualization and cadence patterns are in section 5.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |