# Palavue prompts and workflows

Ready-made prompts for an assistant that has the Palavue connector or API key. Replace places and measure ids as needed.

## Look up a place

**One county profile**

> Use the Palavue tools. Search for "Nantucket" in MA, get its FIPS code, pull the county, and explain each measure in plain words, using the measure definitions. Say which values are estimates or unavailable.

**A state at a glance**

> Pull every county in Colorado with the Palavue pull tool, only measures NC05 and NC14. List the five highest and lowest by national_rank for each, with the unit.

## Compare and find outliers

**Compare two places**

> Compare Park County, CO and Borden County, TX on every measure with a reported value. Tabulate value, rank and cohort_n. Do not compare ranks across measures without noting that cohort_n differs.

**Find the unusual counties**

> Pull every county, only measure NC05. Find the 15 counties where pay is furthest above peers and the 15 furthest below. Skip rows whose status is not reported. Say what the measure means and what it cannot say.

## Write a credited briefing

**One-page briefing**

> Write a one-page briefing on the economy of Kings County, NY from the Palavue data. Use only pulled values. End with the credit line: "Source: Palavue County Economy Bundle (palavue.com), edition <edition>." Include the data caveats.

## Check quality and caveats

**Audit a pull**

> For the counties I pulled, count rows by status (reported, estimated, unavailable, held) per measure. Report which measures have the most gaps and read their blank_means and cannot_say text from the measure definitions.

**Before drawing a conclusion**

> Before telling me what drives a result, list what each measure used cannot say. Remember a rank or association is not a cause and an estimate is not a forecast.
