Apps surface

Operational Data Explorer for manufacturing panels and live site telemetry

This sprint turns the explorer into a real workbench instead of a notebook artifact. The app now defaults to a live manufacturing dataset, can pivot into your own site metrics, and stays visually aligned with the rest of the apps host.

Live manufacturing panel from FRED / Census M3 Latest observation: 2026-08

Jump to explorer Data Explorer Metrics System

Interactive explorer

The explorer is embedded directly into the page so it behaves like a first-class surface, not an iframe bolt-on. Use it to compare manufacturing signals, inspect runtime telemetry, and generate a clean notebook handoff for deeper analysis.

Data modes and workflow support

The explorer is the primary surface. These supporting cards stay full-width below it so the charts keep their aspect ratio while the mode and handoff context remains close by.

Available data modes

The app can shift between manufacturing, site telemetry, and fallback sandbox modes without changing the analysis workflow.

Local Sandbox Synthetic fallback dataset

Live Manufacturing FRED / Census M3 manufacturing panel

Site Metrics Production telemetry from SystemMetricSnapshot

Why this layout works better

The support layer now reads as documentation and workflow context rather than competing with the live analytical surface.

Notebook-friendly flow The handoff block mirrors the current slice in Python so you can move between app and notebook without restarting the analysis from scratch.

Metrics companion The site metrics adapter lets the same analysis workbench inspect operational telemetry, not just manufacturing data.

Stateful without cross-talk The new Dash runtime keeps interaction state in stores and callbacks instead of a shared singleton, which avoids user bleed-through and makes the layout much more predictable.

Examples worth trying

The Example playbook menu inside the explorer applies complete analytical slices instead of leaving visitors to guess at useful feature combinations.

Yield rescue Ranks controllable process or market drivers against the active yield or shipment target, then shows whether the model has enough signal to trust.

Inventory pressure Compares lagging demand and capacity signals so the trend, relationship map, and handoff code all answer the same operating question.

Handoff-ready slice Builds a compact, reproducible Python packet for follow-up notebook work with the current adapter, date window, categories, target, and features preserved.