Engineering notes: how the demos are built and how we will measure
How the demonstrators on this site are built today, what a production version would add, and how we intend to measure cost and performance before publishing any numbers.
How the demonstrators are built today
The demos are deliberately simple so that what you see is exactly what runs. There is no server-side component and no connection to Microsoft Fabric.
| Part | Implementation |
|---|---|
| Pages | Static HTML generated at build time from content files. Every page has its own URL, metadata and structured data. |
| Demo logic | Plain JavaScript modules running in the browser, one per demo, sharing a small core for storage, validation messages and event tracking. |
| Sample data | Generated deterministically from a fixed seed, so every visitor starts from identical figures. The same data modules render the static previews on the home page. |
| Changes | Saved in the browser’s local storage under a versioned key per demo. “Reset demo” deletes the key and restores the sample data. |
| Calculations | OEE, loss breakdown and financial variances are calculated in the browser. The formulas are shown on each demo page. |
| Network | Demos make no network requests. Photos in the quality demo are resized in the browser and never uploaded. |
| Analytics | Named events (demo_open, demo_interaction, demo_reset and others) are pushed to window.dataLayer without form content or personal data. No analytics tool is connected yet. |
Keeping the numbers consistent
A demo that shows inconsistent numbers teaches the wrong lesson. Two rules hold across the prototypes:
- OEE is always Availability × Performance × Quality. All three are derived from the same underlying minutes and counts. When several lines are combined, quality is weighted by ideal cycle time, so the product equals fully productive time divided by planned time at every level. Downtime events always add up to the downtime used in availability.
- Variance assessment depends on account type. Variance is Actual − Budget for every account. A positive variance is favourable for revenue and unfavourable for cost. The assessment is shown as text and an arrow, so it does not rely on colour.
Reference architecture for a production solution
Each demo page describes a proposed production architecture for its use case. They share a common shape:
- Sources ERP, MES, planning or quality systems remain the systems of record.
- Ingestion and storage Data lands in OneLake items (lakehouse or warehouse) on a schedule that matches the decision cadence.
- Semantic model KPIs such as OEE or variance are defined once and reused by reports and the application.
- Write-back store User input (actions, issues, commentary) is stored with status, owner and full change history.
- Application layer Focused views for the workflow, with permissions from Microsoft Entra ID.
- Reporting on input Captured input flows back into reports, so the effect of decisions can be analysed.
Architecture guidance is a proposed design, not a deployed reference implementation. Validate platform capabilities, authorization and consumption in your own tenant. See Microsoft’s Fabric Apps overview.
How we will measure cost and performance
We will publish measurements only when they come from a documented test on a real Fabric environment. Until then this section describes the method, so that results can be checked when they appear.
| Metric | Definition |
|---|---|
| View load time | Time from opening an app view to all figures shown, median and 95th percentile. |
| Write confirmation time | Time from a user saving input to the confirmed write, median and 95th percentile. |
| Input-to-report latency | Time until saved input is visible in the related report. |
| Capacity consumption | Capacity used per 1,000 scripted interactions, read from Microsoft’s capacity monitoring tools. |
| Cost per scenario | Capacity consumption converted to cost using the published price for the region and capacity size at the test date. |
Test protocol
- Three dataset sizes per demo (small, medium, large), with row counts stated.
- Scripted user journeys that mirror the demo workflows, run with a stated number of concurrent users.
- Cold and warm runs reported separately, with at least ten repetitions each.
- Environment recorded with every result: date, region, capacity size, item types and configuration.
- Test scripts and raw results published alongside the summary.
Architecture guidance is a proposed design, not a deployed reference implementation. Validate platform capabilities, authorization and consumption in your own tenant. See Microsoft’s Fabric Apps overview.
No measurements published yet
No cost or performance results are available. They will appear on this page and in Updates with the date and environment of the test.