Semantic Layer AI Experiment Reproduction Files

Prerequisite:
Use the six-table synthetic dataset from the earlier Databricks Genie experiment:
customers, orders, order_items, products, returns, sales_targets
in workspace.genie_sales.

Run files 01 through 05 in order.
Run file 06 and confirm:
- Q2 return rate displays as about 4.11%
- Central Q2 target shortfall is 241,279.86
- Active customer count as of 2026-08-31 is 285

Create a new Genie Agent with exactly these metric views:
- workspace.genie_sales.sales_semantic_metrics
- workspace.genie_sales.regional_performance_metrics
- workspace.genie_sales.customer_activity_metrics

Paste 07_round_c2_instructions.txt into the general instructions.
Keep curated examples at zero for the controlled benchmark.
Run the 14 questions in Chat benchmark mode.

Observed result in this experiment:
- Databricks official Chat benchmark: 13/14, 93%
- Manual business review: 14/14, 100%
- The one official rejection was a representation mismatch:
  model result 0.041064 equals 4.1064%, while the ground-truth result was stored as 4.11.

This is a controlled synthetic experiment, not a universal accuracy claim.
