nuMetrix: SZO Material Analytics
Demo Walkthrough — March 2026
48,382 materials · 68,441 cases · 2.27M billing events
Real data from Spitalzentrum Oberwallis (Brig/Visp)
What you're about to see
The system analyzes SZO's real data through multiple diagnostic lenses.
Every hypothesis with a confirmed verdict also has a diagnosis — a structured root cause with confidence score and recommendation. The system doesn't just count anomalies; it explains them.
Live demo step
Open the Explorer at localhost:4000. With only zeta built, it auto-redirects to hospital_zeta. Show the overview page with its summary cards.
Let's zoom into one specific story.
The Catalog Maintenance Gap
Hypothesis: is_active flags and validity dates are drifting apart
Four probes feed into this hypothesis — all four fired:
| Probe | Role | Findings |
|---|---|---|
| Active but date expired | Primary | 1,368 |
| Inactive but date valid | Primary | 1,413 |
| Stale articles | Supporting | 964 |
| Catalog health assessment | Context | 3,135 |
Live demo step
Navigate to Hypotheses → click "Catalog Maintenance Gap". Show the evidence chain and confirmed verdict.
Ronnie's Question
The question that changed the probe.
This is domain knowledge that no data model can derive. The system knew something was wrong, but Ronnie knew what to do about it.
Activity-Based Classification
Ronnie's rule: purchasing activity tells you which field is wrong.
| Probe | Has Purchases? | Likely Wrong Field | Action |
|---|---|---|---|
| Active + date expired | Yes | Date | Extend validity — article genuinely in use |
| Active + date expired | No | Flag | Set inactive — article genuinely expired |
| Inactive + date valid | Yes | Flag | Re-activate — article genuinely in use |
| Inactive + date valid | No | Date | Shorten validity — article genuinely inactive |
Simple rule, massive impact. This logic is now built into the probes — every finding carries the classification automatically.
What the data says
Real SZO numbers after applying Ronnie's classification.
Active + Date Expired (1,368)
| Activity | Wrong Field | Count | CHF |
|---|---|---|---|
| Purchases | Date | 610 | 18,984 |
| None | Flag | 758 | 92,706 |
65 findings escalated to high severity (active + purchases + CHF>1K = urgent date fix)
Inactive + Date Valid (1,413)
| Activity | Wrong Field | Count | CHF |
|---|---|---|---|
| Purchases | Flag | 216 | 14,860 |
| None | Date | 1,197 | 109,345 |
16 findings escalated to high severity (inactive + purchases = being bought under the radar)
Live demo step
Navigate to Findings → select probe_active_date_expired. Expand a finding to show the has_purchases, likely_wrong_field, and purchase_count evidence fields.
How did we get from conversation to code?
Knowledge Capture: The SMEbit
Ronnie's insight — formalized, attributed, versioned.
Activity-based classification of is_active/date contradictions
When an article's is_active flag contradicts its validity date range, purchasing activity in the observed period reveals which field is likely wrong. Active articles with recent purchases have a wrong date; active articles without purchases have a wrong flag.
Live demo step
Navigate to SMEbits → find "Activity-based classification of is_active/date contradictions". Show the provider attribution, anchors, and the "why" field.
The Knowledge-Driven Workflow
How expert knowledge becomes system intelligence — in one session.
Capture
Ronnie shares his classification rule. It becomes an SMEbit — attributed, scoped, versioned.
Consequence Analysis
Which probes need to change? What evidence fields to add? How does severity shift?
Implement
Probes are enriched with billing/usage joins. Activity classification built into every finding.
Rebuild & Review
Full rebuild across all 5 tenants. Same findings, richer evidence. 81 findings escalated to high severity.
From conversation to production: under 30 minutes. The knowledge is permanent — it survives every rebuild, every new tenant.
The Bigger Story: BitBundles
SMEbits accumulate. BitBundles weave them into narratives.
Material Data Quality at SZO
A curated narrative connecting 5 SMEbits about material master data quality at SZO — from wrong descriptions to validity contradictions to catalog maintenance gaps.
Each SMEbit is atomic — one insight, one provider, one scope. BitBundles are the story layer: "Here's what we know about material data quality at this hospital, and here's why it matters."
Live demo step
Navigate to BitBundles → open "Material Data Quality at SZO". Show the narrative and the linked SMEbits.
Ronnie has four more questions.
Catalog Inventory KPIs
Ronnie asks four questions. Each one is now a probe — finding count IS the answer.
| Ronnie's Question | Probe | Answer (SZO) |
|---|---|---|
| How many unique active articles? | probe_active_materials | 22,402 |
| How many unique inactive articles? | probe_inactive_materials | 25,980 |
| Active articles from a single vendor only? | probe_active_single_source | 487 |
| Articles delivered by multiple vendors? | probe_multi_source_materials | 717 |
All four feed into assessment_catalog_inventory (48,382 materials assessed). The same pattern: question → probe → instant answer across any tenant.
Live demo step
Navigate to Findings → select probe_active_materials. The finding count in the header is Ronnie's answer: 22,402 active articles. Switch to probe_active_single_source to show the 487 single-vendor materials with supplier names and billing volumes.
The Report
Ronnie's questions — answered, explained, printable.
| Bucket | Count |
|---|---|
| Active, single supplier | 487 |
| Active, multiple suppliers | 717 |
| Active, no supplier data (derived) | 21,198 |
| Total active | 22,402 |
Every number links to its probe. Every probe shows the SQL logic. Print to PDF or share directly.
Live demo step
Navigate to Executive Summary (sidebar) → click Catalog Inventory. Show the business questions table, supplier breakdown, and lineage section. Hit Ctrl+P to show the print-ready PDF.
Search for "Ronnie"
The JumpBar finds people, probes, hypotheses, materials — everything in one search.
Live demo step
Press Cmd+K (or J) to open the JumpBar. Type "ronnie". Show the People result with contribution counts.
The big picture.
Confirmed Hypotheses at SZO
10 out of 11 hypotheses confirmed by evidence.
| Hypothesis | Diagnosis | Flagged Volume (CHF) |
|---|---|---|
| Cost center misallocation | data quality | 66.7M |
| Cross-site leakage | process failure | 66.5M |
| Stale catalogue | process failure | 39.2M |
| I/O imbalance | process failure | 35.4M |
| Revenue leakage (unbilled) | process failure | 34.8M |
| Duplicate billing | data quality | 33.3M |
| Supplier concentration risk | structural | 28.7M |
| Overpriced generics | structural | 28.4M |
| Catalog maintenance gap | process failure | 6.7M |
| Phantom cases | data quality | 33.3M |
Each confirmed hypothesis has a diagnosis with root cause category, confidence score, explanation, and recommendation. The system doesn't just flag — it explains.
Live demo step
Browse the Hypotheses page. Click into one (e.g., "Stale Catalogue") to show the diagnosis section with root cause and recommendation.
The Analytical Pyramid
From raw data to actionable root causes — every layer adds meaning.
What happens next?
Every conversation adds knowledge. The system grows smarter with each SMEbit.
What Ronnie can do now
- Share domain rules → they become SMEbits
- Challenge probe results → probes get sharper
- Curate BitBundles → build the institutional narrative
- Validate findings → calibrate confidence
What the system does
- Attributes every insight to its provider
- Versions everything — knowledge evolves
- Runs across all tenants — portable rules
- Rebuilds in minutes — instant feedback
Thank you
nuMetrix · Hospital Material Flow Analytics
55 SMEbits · 15 BitBundles · 38 Probes · 11 Hypotheses · 10 Diagnoses
All knowledge attributed. All results reproducible.