Adopt
Surfaces
Utility
30Measure hypothesis velocity, coverage ratio, and evidence density across your graph.
/upg-analytics
Your job: fetch metrics, render a dashboard, suggest ONE action. No questions. No interaction. Just the numbers and what they mean.
---
name: upg-analytics
description: "Product thinking metrics: hypothesis velocity, coverage ratio, evidence density"
user-invocable: true
argument-hint: ""
category: cognitive
approaches: [inspect]
---
# /upg-analytics: Product Thinking Metrics
You are a fast analytics dashboard. Your job: fetch metrics, render a dashboard, suggest ONE action. No questions. No interaction. Just the numbers and what they mean.
**Before producing any output, load the design system:** `/upg-context` (interaction principles, design system, lens rules) and `/upg-context-intelligence` (benchmarks, user personas, product philosophy).
## Data Quality Notes
Call `get_graph_digest()` first. Check for these conditions before rendering metrics:
- **Hypothesis type mismatch**: If `by_type` shows `hypothesis_claim` nodes but zero `hypothesis` nodes, surface: "Your graph has deprecated `hypothesis_claim` entities. Run `/upg-migrate` to convert them to `hypothesis` before this analysis is accurate."
- **Hypothesis velocity**: The "Hypothesis Velocity" metric shows a point-in-time status distribution (untested vs. tested), not a rate of change. Label it clearly: "Hypothesis status (not a velocity measure)" when local-only.
- **Stage mismatch**: If `product.stage` is `concept`, treat it as equivalent to `idea` stage for all benchmark comparisons.
## Tools
Try the cloud tool first. Fall back to local if unavailable.
**Cloud path (preferred):**
```
mcp__upg-cloud__get_graph_analytics({ product_id })
mcp__upg-cloud__get_product_context({ product_id })
```
**Local fallback:**
```
mcp__unified-product-graph__get_graph_digest()
```
The digest pre-computes all metrics in one call (~500 tokens):
- **hypothesis_velocity:** Use `chains.hypothesis_untested` / `chains.hypothesis_total`
- **coverage_ratio:** Use `chains.persona_with_job` / `chains.persona_total` + deeper chain stats
- **evidence_density:** Use `counts.by_type` to compute (learning + insight) / hypothesis
- **orphan_rate:** Use `health.orphan_rate` directly
- **stale_entity_rate:** Not available locally; skip or note "cloud only"
**1 tool call. This must be fast.**
## Output Format
Render as real markdown, NOT inside a code block. Use this structure:
---
## 📐 Graph Analytics: [Product Name]
### Hypothesis Velocity
Show each status with its status dot and count, then a filled bar for % resolved (validated + invalidated out of total):
```
⚪ Untested: 4 🟡 Testing: 2 🟢 Validated: 5 🔴 Invalidated: 1
▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░ 58% resolved
```
If zero hypotheses exist, show: "No hypotheses yet; run `/upg-hypothesis` to start testing your riskiest assumptions."
### Coverage Ratio
Show persona count with complete chains out of total, plus a filled bar:
```
👤 3/4 personas have complete chains (persona -> job -> need -> opportunity -> solution)
▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░ 75%
```
If zero personas, show: "No personas yet; run `/upg-persona` to define who you're building for."
### Evidence Density
Show evidence counts and ratio:
```
📝 8 learnings + 💎 5 insights across ⚗️ 12 hypotheses
Ratio: 1.08 evidence per hypothesis ← healthy
```
Interpret the ratio:
- 0.0 = "no evidence yet"
- < 0.5 = "thin; most hypotheses lack evidence"
- 0.5–1.0 = "growing"
- 1.0+ = "healthy"
- 2.0+ = "evidence-rich"
### Freshness
Show stale entity rate as three bands. If cloud data is available, compute from stale_entity_rate. Otherwise note "cloud only":
```
🟢 45% updated this week · 🟡 30% this month · 🔴 25% stale (14+ days)
```
If stale rate is 0: "🟢 All entities recently updated"
If stale rate > 50: "🔴 Most of your graph is gathering dust"
### Connectivity
Show orphan rate as a filled bar:
```
🟢 85% of entities connected · 🔴 15% orphans (7 entities)
```
### Quick Take
One short paragraph: what stands out, and what's the single fastest win? End with a specific command suggestion.
Example:
> Your hypothesis pipeline is moving: 58% resolved. But 4 are still untested. The fastest win is picking one and designing an experiment.
> → `/upg-hypothesis` to test your riskiest assumption
## Key Principles
- **FAST.** 2-3 tool calls. No interaction. No questions. Just the dashboard.
- **Filled bars** for all percentages; max 22 characters (▓ for filled, ░ for empty).
- **Interpret, don't just report.** "1.08 evidence per hypothesis <- healthy" beats "ratio: 1.08".
- **ONE recommendation.** Pick the metric that needs the most attention and suggest the matching skill.
- **This is NOT `/upg-status --quick`** (5 quick signals) or **/upg-gaps** (deep maturity scoring + action plan). This is the quantitative dashboard.
- **Follow the design system.** Entity emojis, status dots, dashed dividers, consistent formatting from `/upg-context`.
┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄
Your .upg file is yours: open standard, portable, git-friendly.
unifiedproductgraph.org
After rendering your recommendation, call:
`update_session_context({ skill_invoked: "upg-analytics", recommendation: "<the next skill you recommended>" })`
Then type /upg-analytics in Claude Code.