Overview
A safe place to break the model — before it breaks a community
The Observability Lab is an in-browser sandbox where students and researchers rehearse the full lifecycle of an algorithmic incident: spotting drift, triaging a fairness regression, and choosing an intervention under real-world constraints.
Every session is instrumented so learners can review their decision path afterwards — no accounts, no setup, no cost.
Key Benefits
Why the Lab matters
Hands-on rigor
Move beyond slideware — students touch the same tradeoffs practitioners face in production.
Reusable curriculum
Scenarios, workbooks, and slide decks drop straight into an undergraduate or graduate syllabus.
Zero setup
Runs entirely in the browser so a full classroom can start a scenario in under a minute.
What to Expect
Your first sandbox session
- 1
Pick a workshop scenario
Load a pre-built failure mode drawn from real municipal deployments.
- 2
Adjust levers and observe
Tune model thresholds, retraining cadence, and oversight rules; watch fairness and accuracy respond in real time.
- 3
Debrief with the trace
Review the structured log of your decisions to reason about tradeoffs and share findings with peers.
Capabilities
Data Drift Simulation
Inject distribution shifts and watch fairness, accuracy, and calibration diverge in real time.
Red-Team Scenarios
Pre-built workshop cases that walk students through failure modes drawn from real municipal deployments.
Diagnostic Traces
Structured logs of every intervention so learners can audit their own decision path and reason about tradeoffs.
Classroom Toolkit
Downloadable workbook, slide deck, and NIST-aligned risk matrix designed for undergraduate and graduate coursework.