Dataset Structure for Template Code Skills

Data Assumptions

Dataset Properties

Query row limit: LIMIT added to each SQL query. A null query row limit will default to a 100 row limit
Dataset start / end date: not automatically generated. Required to hit outside of time range guardrails. If not set, the analysis will hit a no data guardrail in these cases instead
Data Interval: Lowest level time granularity
Source: A dataset can only be created from a single table or view Metrics
Assumes metrics are their own columns in the table.
Assumed to be a numeric type
Assumed to have an aggregation function
Can include null values, but will show up as NaN in the analysis

Dimensions

Date Dimensions

Can't include null values.
Assumed to be discrete values, not time ranges.
Assumes date dimensions are in YYYY-MM-DD format
Requires columns to be in the following format, max\time{time\granularity}, up from the lowest time granularity available (set by the Data Interval).
- Supported time granularities
  - date, week, month, quarter, year
- For example, if the dataset is daily data and you'd like to aggregate across weeks, months, quarters, and years, you'd have the following columns:
  - max_time_date, max_time_week, max_time_month, max_time_quarter, max_time_year
Dimension Properties
- Should be hidden dimensions so the LLM doesn't use them explicitly during interpretation.
- Requires Data Type set to date
- Requires the SQL and SQL Sort Expression set
  - SQL is used to show the output values of the date at that granularity, like W15 2024, Jan 2025, Q1 2023, 2021
  - SQL Sort Expression is used to order the dates in ascending order.

max\time_date (Assumed to be a date type)

max\time_week

SQL: 'W' || EXTRACT('week' FROM <your_date_field>) || ' ' || EXTRACT('year' FROM <your_date_field>)
SQL Sort Expression: DATE_TRUNC('week', CAST(<your_date_field> AS DATE))

max\time_month

SQL: LPAD(EXTRACT('month' FROM <your_date_field>)::VARCHAR, 2, '0') || '/' || EXTRACT('year' FROM <your_date_field>)
SQL Sort Expression: DATE_TRUNC('month', CAST(<your_date_field> AS DATE))

max\time_quarter

SQL: 'Q' || EXTRACT('quarter' FROM <your_date_field>) || ' ' || EXTRACT('year' FROM <your_date_field>)
SQL Sort Expression: DATE_TRUNC('quarter', CAST(<your_date_field> AS DATE))

max\time_year