[mattsenicksigma]

Summary

Reports have no date control, so a "last 7 days" breakdown next to an all-time trend has to be built into the element. Today that means hand-rolling a hidden column and filter outside the package (the OpenRouter cost report in dbt-sos does exactly that).

• New filter_last_days(element, column, days): limits one chart, KPI or table to the last N days of a date column, measured from the day the report renders. Adds a hidden If([Col] >= DateAdd("day", -N, Today()), 1, 0) column and a number-range filter keeping 1 (the same filter kind table(window=) already emits), so Today() is evaluated by Sigma at render time and a scheduled report never goes stale. Fails loudly (not silently unscoped) on a non-warehouse element, a bad days, or the same column limited twice. Composes with filter_by_value.
• simple_dashboard: date_column= plus a 'last_days': N key on any KPI, chart or the pivot, so elements are windowed individually and the rest keep full history.
• simple_dashboard: table_pivot={'rows', 'cols', 'values', 'name', 'h', 'last_days'} as an alternative to table_columns, for totals by group rather than row-by-row (mutually exclusive, enforced).
• Documented in schema.yml and the README; new sigma_last_days_dashboard integration model, golden file, assertions, and 6 negative cases.

Default behavior is unchanged: nothing is windowed unless last_days is given.

Test plan

• UPDATE_GOLDEN=1 PARSE_FLAGS=--warn-error bash scripts/run_integration_tests_duckdb.sh on dbt 1.12.5 / dbt-duckdb 1.11.0: all pass (96 negative cases, schema and layout checks); the only new golden is sigma_last_days_dashboard.json
• dbt 2.x CI
• Confirm live in Sigma: not exercised here; the filter kind is the same one windowed tables use, and the compiled spec validates against Sigma's request schema

🤖 Generated with Claude Code