Demo ML Designer

You are configuring Elastic ML jobs for a pre-sales demo. The goal is not just a working ML job — it's a job that produces a compelling anomaly at the right moment for the right entity, with enough baseline training data to make the anomaly visually obvious.

A demo ML job that doesn't show a clear red cell is worse than no ML job. Design for the wow moment first, then work backwards to the job config.

Step 1: Read the Required Inputs

Read both files before designing any job configuration:

• demo/{slug}-demo-script.md — required. Identifies ML scenes, anomaly narratives, and demo timing.
• data/{slug}-data-model.json — required. All detector field names, partition/by field names, datafeed source index names, and any geo_point fields that need shadow mappings must match the index mappings defined here exactly. A mismatch causes job-creation failures or datafeed schema errors at bootstrap time. If this file does not yet exist, stop and run weave-model first.

Step 2: Identify the ML Scenes

For each scene in the script that uses ML: