DataCEVA US Layoff Notices

View saved version →

---
name: company-layoff-brief
description: "Use for questions about a specific company's U.S. layoffs, WARN Act notices, workforce reductions, plant closures or layoff history: whether a company is laying people off, an employer's layoff or WARN history, or its job cuts, closures or downsizing. Writes a sourced company layoff briefing. Use the DataCEVA US Layoff Notices connector to find and verify notices, counts, dates and state coverage."
---

# Company layoff brief

Write a short, sourced briefing on one employer's WARN layoff notices, using the DataCEVA US Layoff Notices
connector. Before answering for the first time in a conversation, read `references/warn-data-notes.md` for what
the data covers and how to cite it.

## Steps

1. **Find the company.** Call `get_company_layoff_history` with the company name the user gave. Add `states` or
   `since` only if the user asked for them.
   - If it returns no notices but offers `suggestions`, list them and ask which one the user means. Don't
     guess.
   - If `matched_companies` contains names that look like different businesses (for example a franchise or a
     company with a similar name), say which ones you included, and leave out the ones that are clearly
     different.
2. **Read the latest notices.** Call `get_layoff_notice` for the one to three most recent notices in
   `latest_notices`, to get the site, dates, worker count as published, layoff type and reason.
3. **Check coverage.** Call `get_data_coverage` for each state in the history (at most five), so you can say
   whether a state's list is current, partial or missing.
4. **Write the brief** in the format below.

## Format

**Headline** (two or three sentences): number of notices and workers, the period they span, the states, and
whether the most recent notice is recent (within about 90 days) or old. Give the data's as-of date.

**By year** (table: year, notices, workers) and **by state** (table: state, notices, workers), from the tool's
`by_year` and `by_state`.

**Latest notices** (table, newest first): filed date, layoff date, site (city, state), workers, type, source
(link to `notice_url`, or the agency's `state_warn_page` when there is no notice link).

**What this does and doesn't show** (three to five bullets), chosen from the notes that apply:
- WARN covers larger layoffs (generally 50 or more workers at one site), so smaller cuts won't appear.
- Totals leave out update and correction notices for an earlier notice; say how many were left out if the tool
  reports it.
- Any state in this history whose list is partial, saved-history-only or unavailable.
- Counts are as filed and can change in later updates.

## Rules

- Use only what the tools return. Don't add numbers, reasons or dates from memory, and don't predict future
  layoffs.
- Keep filed date and layoff date distinct, and label which one a date is.
- Never describe or look for individual employees.
- If the company has no notices, say that no WARN notices were found in the covered states, not that it has
  had no layoffs.
- Keep the brief under about 400 words plus tables unless the user asks for more.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: layoff-trend-analysis
description: "Use for U.S. layoff trend research, comparisons, rankings and statistics from WARN Act notices: whether layoffs are rising or falling, which states, industries or companies filed the most notices, and how time periods, states, census regions, industries or companies compare by notices or affected workers. Use the DataCEVA US Layoff Notices connector to retrieve comparable data and check state coverage before drawing conclusions."
---

# Layoff trend analysis

Answer trend and ranking questions with WARN layoff notices, using the DataCEVA US Layoff Notices connector,
and make every comparison fair. Before answering for the first time in a conversation, read
`references/warn-data-notes.md` for what the data covers and how to cite it.

## Steps

1. **Define the comparison.** Decide what is compared (periods, states, regions, industries or companies), the
   measure (notices, workers, or both) and the period. With no period, use the last two full years by month,
   or the last five full years by year for long-run questions.
2. **Check coverage first.** Call `get_data_coverage` (for the states involved, or all states for national
   questions). For each state, note `first_filed`, `last_filed`, `last_refreshed` and `status`. Only compare
   periods that every included state covers. Leave out, or list separately, states whose history starts later,
   ends earlier, or is unavailable.
3. **Get the numbers.** Call `summarize_layoffs` with the matching `group_by` (`month`, `quarter`, `year`,
   `state`, `census_region`, `industry` or `company`) and filters.
   - If GA, NJ, PA or MI matter to the question, run it a second time with `include_scheduled_only: true` and
     explain that those states publish only planned layoff dates.
   - For industry questions, say that industry is reported by some states only (the tool's caveats name them).
4. **Make it fair.**
   - Treat the current period as incomplete: label it, or compare like with like (for example January to
     September of each year).
   - Show changes as numbers and percentages, and don't call small differences a trend.
   - Rankings: give both notices and workers; one very large notice can dominate workers.
5. **Write the analysis** in the format below.

## Format

**Answer** (two or three sentences) that directly answers the question with the key numbers.

**Table** of the series or ranking (period or group, notices, workers, change).

**How this was measured** (two to four bullets): the period, the states included and left out, filed date or
planned layoff date, and that update notices are not counted twice.

**Limits** (bullets that apply): WARN covers larger layoffs only, so it is a floor rather than a count of all job
losses; industry coverage; partial or missing states; the as-of date.

## Rules

- Use only what the tools return; don't mix in other statistics (such as BLS figures) unless the user provides
  them, and label any outside figure as not from WARN data.
- Don't forecast future layoffs; describe past and current filings only.
- Never describe or look for individual employees.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: local-layoff-scan
description: "Use when someone asks about layoffs near a U.S. city, ZIP code, state, region or location: which employers near a place filed WARN layoff notices, recent local layoffs and plant closures, notices within a radius, what happened in an area this week, month or quarter, or how it compares with the period before. Use the DataCEVA US Layoff Notices connector to search notices, summarize affected workers, compare periods and check data coverage."
---

# Local layoff scan

Summarize WARN layoff notices for one place and period, using the DataCEVA US Layoff Notices connector. Before
answering for the first time in a conversation, read `references/warn-data-notes.md` for what the data covers and
how to cite it.

## Steps

1. **Pin down the place and period.**
   - Place: a city as `City, ST`, a 5-digit ZIP, a state, or a census region (Northeast, Midwest, South, West).
     The data covers U.S. states and D.C. only; for anywhere else, say so and stop.
   - Radius: 25 miles unless the user says otherwise (up to 100).
   - Period: turn "this month", "last quarter" and similar into `filed_after` / `filed_before` dates. With no
     period, use the last 90 days and say so.
2. **List the notices.** Call `search_layoff_notices` with the place (or `states` / `census_region`) and dates,
   `sort: "largest"`, `limit: 20`. If `total_matching` is larger than what came back, say how many more there
   are.
3. **Break it down.** Call `summarize_layoffs` with the same filters and `group_by: "industry"`. For periods
   longer than two months, also call it with `group_by: "month"`.
4. **Compare with the period before.** Call `summarize_layoffs` again with the same filters for the previous
   period of the same length, and compare notices and workers.
5. **Check coverage.** Call `get_data_coverage` for the state (or each state in the region, if few), so you can
   say whether its list is current.
6. **Write the scan** in the format below.

## Format

**Answer** (two or three sentences): how many notices and workers in the place and period, the largest one,
and the change from the period before (as numbers, not just "up" or "down").

**Notices** (table, largest first): company, city, filed date, layoff date, workers, type, industry, source
link (`notice_url`, or the agency's `state_warn_page`).

**By industry** (table: industry, notices, workers), with a note that industry comes from some states only.

**Compared with the period before** (one short paragraph). Don't call a change a trend when the counts are
small (for example 3 notices vs 2).

**Limits** (bullets that apply): approximate locations (ZIP, city or county centre); states that publish only a
planned layoff date (GA, NJ, PA and much of MI) are left out of filed-date results unless
`include_scheduled_only` is set; any state that is partial, saved-history-only or unavailable; the as-of date.

## Rules

- Use only what the tools return; don't add layoffs from memory or the news.
- Keep filed date and layoff date distinct, and label which one a date is.
- Never describe or look for individual employees.
- If nothing is found, say no WARN notices were filed in the covered data for that place and period, not that
  there were no layoffs.

Referenced files: 1 — file references are metadata, not package contents.