The Layoff-Filing Teardown: One Notice to a Team-Level, Dated Slate
From one state layoff filing, estimate which teams were cut, grade each estimate, split notice-period from separated people, and date every outreach segment.
Key takeaways
- Microsoft's May 2025 Washington filings listed 817 software engineering, 373 product management, 218 technical program management and 55 business program management positions, enough detail to test the company's own account of the cuts.
- Microsoft framed its May 2025 cuts as removing managers, but only 17% of those affected were managers, so grade any company statement as inferred until title counts or named posts back it up.
- In Refolk's index, PM and TPM profiles are 9.9% of the US Microsoft-headline engineering and program population but were 42.0% of the filed SWE plus PM/TPM cuts, about 6.6x over-represented.
- In a 25-profile sample from Refolk's index, 9 of the US software engineer profiles with 'ex-Microsoft' in the headline still listed Microsoft as their current employer, so source from the current-employer field, not the headline.
- People in a paid notice period are still employees: Google's Washington round notified staff between July 8 and August 4 but separated them on September 6, September 13 and October 4.
- A short-form notice under 20 CFR 639.7(f) carries no job titles, and a multi-site cut can produce no filing at all, as with Pluralsight's layoff of 400+ people.
This guide is for in-house recruiters, sourcers and founders who have just seen a company on their source list file a layoff notice. It takes one real filing, Microsoft's May 2025 Washington notice, and carries it all the way to a dated outreach plan: matching the filing entity to the company, decoding title counts into teams, grading each estimate, separating people still on payroll from those already gone, and clearing the legal questions. The wrong turns are left in, because they are where most layoff sourcing goes wrong.
What does a state layoff filing tell a recruiter?
A full state layoff notice tells you the site, the separation dates, and the job titles of affected positions with a count for each title. A short-form notice tells you only the first separation date and a headcount. Which one was filed, and whether the state publishes the titles, decides how much team-level work you can do.
A WARN notice is the advance layoff notice an employer files with the state dislocated worker unit when a cut crosses the federal or state threshold. Under 20 CFR 639.7(e), the full notice must list:
- the site of the layoff
- whether the cut is permanent or temporary
- the expected date of the first separation and the separation schedule
- the job titles of positions to be affected, and the number of affected employees in each job classification
- whether bumping rights exist
The short alternative under 639.7(f) needs only the expected date of the first separation and the number of affected employees. No titles. The employer must keep the rest of the 639.7(e) information on site, so the titles exist. They just are not public.
States also differ in what they publish. This is the first thing to check, because it tells you in five minutes whether title-level decoding is possible at all.
| State portal | Titles public? | What to expect |
|---|---|---|
| California EDD | Yes, by location if multi-site | XLSX report |
| Washington ESD | Yes, in the notice PDF | Notice copy without employee information |
| New York DOL | Collected at filing; public display unverified | Dashboard export: date, site, county, reason, type |
| Texas TWC | Not established | Excel files only |
| New Jersey | Not established | Dates at month precision only |
| Georgia TCSG | Yes | Public view lists titles |
For New York, open a dashboard export for a company you know and see whether a title column appears. Do not assume either way.
The worked filing: matching Microsoft's notice to sites and siblings
The worked example is Microsoft's May 2025 Washington filing, which tied 1,985 positions to Redmond and carried title-level rows because Washington publishes the notice PDF. Before any decoding, record the entity, the sites, the dates and the scope, then check whether the filing is the whole cut or one slice of it.
Reading the filing
From the notice, record five things: the legal entity name, every site address, the notice date, the separation date or schedule, and whether title counts are present. For the Microsoft filing, the sites read "Redmond and Bellevue", and the title rows were present. That second fact matters most: it is why Bloomberg could count titles in the Washington filings and why this teardown is possible.
Matching the entity
Filings use legal names, which can differ from the brand or a DBA (a "doing business as" trade name). A subsidiary or a differently named legal entity can make a filing invisible to a search on the brand name. WARNTracker keeps a "Company Name in Raw Data" column for exactly this, so you can see the name as the state recorded it. Done means every site on the notice maps to a named business unit or office.
Checking siblings
One filing is rarely the whole layoff. Microsoft's worldwide cut was about 6,000 to 7,000, so the 1,985 Washington positions were somewhere between roughly 28% and 33% of it. Search the other state portals and the 90 days either side of the notice for sibling filings. Google's Washington notice is a good model of what a multi-site sibling looks like: one notice covering Kirkland, Seattle, Redmond and remote workers.
Wrong turn 1: splitting by site. The tempting move is to split 1,985 between Redmond and Bellevue and assign each half to whatever teams sit in each building. The filing gives no per-site title rows, so any split is invented. Keep multi-site counts whole unless the filing breaks them out by location, as California requires for multi-site layoffs.
Microsoft May 2025: from worldwide cut to decodable rows
- about 6,000 to 7,000Worldwide cut
Press figure, not a filing
- 1,985Washington filing, Redmond and Bellevue
Title rows public
- 1,463SWE, PM, TPM and BPM rows
Sum of four reported titles
- 591PM plus TPM rows
The over-represented group
The four reported function rows sum to 1,463. If the 1,985 figure and the title counts describe the same filing set, about 522 positions sit in titles the reporting did not break out. The two numbers come from different write-ups, so confirm against the PDF before you rely on that remainder.
Which teams were cut: decoding job-title counts into buckets
Group each filed title into a function, sum each function, and compare its share of the cut with its share of the company's visible population. The comparison, not the raw count, tells you which teams were targeted.
The Microsoft rows were: software engineering 817, product management 373, technical program management 218, business program management 55. Read raw, software engineering is the biggest number, and the early headlines said coders were hit hardest. Now set the same functions against the population visible in Refolk's index of professional profiles.
| Function | Filed WA count | Index US headline profiles | Filed share | Index share |
|---|---|---|---|---|
| Software engineering | 817 | 22,208 | 58.0% | 90.1% |
| PM + TPM | 591 | 2,439 | 42.0% | 9.9% |
| SWE to PM/TPM ratio | 1.38 | 9.11 |
Shares are of the SWE plus PM/TPM total only; business program management is left out because the index count was not pulled for it.
Wrong turn 2: reading the largest row as the target. Software engineering was the largest row and still under-represented against its population. Product and program roles took a share of the cut far beyond their share of the visible workforce. If you are hiring TPMs, this filing was more relevant to you than the headlines suggested.
Wrong turn 3: taking the company's framing. Microsoft described the cuts as reducing management layers. The title counts showed only 17% of those affected were managers. A company's framing is a claim, and the filing is how you test it.
Two limits on this table. The index counts match headline keywords, not a verified employer roster, so the population side is an approximation. And a function is not a team: "product manager" does not tell you whether the person worked on cloud, developer tools or anything else. Function buckets are where decoding stops being confirmed and starts needing other evidence.
Grading each team estimate: confirmed, corroborated, inferred
Grade every team-level claim by its best evidence: confirmed if the filing states it, corroborated if the filing plus press or a named public post agree, and inferred if it rests only on profile movement, anonymous posts, or the company's own statements. The grade goes on the sheet next to the estimate and decides how specific your outreach can be.
Evidence ladder for team attribution
- Filing title rowsConfirmed. Function and count stated in the notice
- Press reporting or named public postsCorroborated. Filing plus an independent, attributable source
- Profile movement and anonymous postsInferred. Pattern without attribution
- Company statements aloneInferred. A claim to test, not evidence
How the grades played out in practice:
- Microsoft, function level: 817 software engineering positions. Confirmed, because the filing states it.
- Microsoft, team level: trade press attributed the cuts to specific teams within days of the filing. Corroborated, where press and filing agree.
- Named individuals: affected employees confirmed their own exits publicly, for example a TypeScript developer posting on Mastodon. That corroborates one person on one team, not the team's whole cut.
- ServiceNow: unverified Blind posts in October 2025 named quality engineering. That was inferred for about 8 months, until a San Diego filing made software quality engineer the largest single title, with 10. At that point it became confirmed.
Sources disagree on how fast grades firm up: Microsoft's cuts were attributed within days, while ServiceNow's rumours took about 8 months to confirm. Grades are revisable, so date each one. How often early attribution turns out wrong has not been established publicly; the Microsoft case is one documented instance, not a rate.
Finding the people: index counts and the keyword noise check
Build the candidate pool from the current-employer field and the end date of the role, not from "ex-company" keywords in the headline. In Refolk's index, more than a third of sampled "ex-Microsoft" engineer profiles still list Microsoft as their current employer.
Here are the intermediate counts for the Microsoft case, with a 25-profile sample drawn from each query.
| Query in Refolk's index | Total US profiles | Sample at Microsoft now | Sample in WA regions |
|---|---|---|---|
| SWE + "Microsoft" | 22,208 | 17 of 25 (68%) | 9 of 25 |
| PM/TPM + "Microsoft" | 2,439 | 21 of 25 (84%) | 11 of 25 |
| SWE + "ex-Microsoft" | 1,599 | 9 of 25 (36%) | 9 of 25 |
What these counts tell you:
- The "ex-" query lies in a specific way. It returns people who came back to Microsoft and people whose "ex-" pointed to someone else. Neither group was in this cut.
- Washington is only part of the population. Only 9 of 25 sampled SWE profiles and 11 of 25 PM/TPM profiles sat in the Washington regions. A Washington filing reaches a fraction of each pool, which is another reason to look for sibling filings.
- The samples are small. Twenty-five profiles is directional. Use these to decide which filter to trust, not to project headcounts.
The practical fix is to ask for the people by role, place and the date their role at the company ended, then check each result's current employer before tagging it. That is the query shape Refolk takes in plain English, and given the PM/TPM finding above, it is where I would start for this filing.
When to reach out to laid-off candidates, and the legal check first
Split the pool into people still in a paid notice period, who are still employees and may be weeks from separation, and people already separated. Reach separated people first, and treat notice-period people as a separate segment that is timed and legally checked.
The notice-period clock
The notice period is the gap between the notice and the separation date. Thresholds and minimum notice vary by regime:
| Regime | Employer size | Layoff trigger | Notice |
|---|---|---|---|
| Federal | 100+ | 50 to 499 if a third of site, or 500+ | 60 days |
| California | 75+ | 50, no percentage test | 60 days |
| New York | 50 full-time | 25 if a third of site, or 250 | 90 days |
| Washington | 50 full-time | per RCW 49.45 | 60 days |
The statutory minimum is not what you plan around. Plan around the separation schedule in the filing, because real gaps vary:
- Google's four California notices ran from 41 to 62 days, and 3 of 4 gave less than 60.
- Google's Washington notice: employees were notified between July 8 and August 4, with separation dates of September 6, September 13 and October 4.
- In Google's January 2023 round, people stayed employed with full payroll and benefits through a paid 60-day notice period, followed by 16 weeks of severance plus 2 weeks per year of service.
- WARNTracker shows several Microsoft Washington rows where the notice date equals the layoff date. That is consistent with pay in lieu of notice, but the reason is unverified.
So tag every person with an estimated end date from the schedule, and replace it with the date they tell you. Reaching people inside the notice window catches them before they post publicly. Whether that improves response is not established: no published comparison of response rates or time to hire exists for layoff-affected versus passive outreach, and nobody has published when recruiter message volume peaks after a filing.
Which segment gets contacted, and when
The legal check
The legal risk sits mainly with your company's relationship to the filer, not with the candidate. An employee's severance non-solicit binds the departing person; one example clause reads "during any Severance Period, he shall not solicit for employment". A company-to-company no-poach clause in an MSA or NDA can bind you as the hirer. Some clauses carve out laid-off staff, for example stating they "do not apply to workers who have been laid off". And an agreement between companies not to "cold call" each other's workers "may constitute an unlawful no-poach agreement".
The procedure: one filing to a dated outreach slate
This is the full sequence, with owner and rough time for each step: about two hours of sourcer work to get from filing to matched sites, a few hours to decode titles, then days of grading and legal review that run in parallel with outreach to already-cleared segments.
Filing to slate
- Find and read the filingSourcer, 30 minutes. Record legal entity, sites, notice date, separation schedule, and whether title counts exist; if not, check whether the 639.7(f) short form was used.
- Match the filing entity to the operating companySourcer, 30 to 60 minutes. Resolve legal-name and DBA differences and map every site to a business unit or office.
- Check other states and the 90-day windowSourcer, 30 minutes. List every sibling filing from the same company, with its sites and counts.
- Decode title counts into team bucketsRecruiter, 1 to 2 hours. Build a table of titles, counts, functions, and filed share versus population share.
- Grade each team estimateRecruiter, 1 to 3 days. Mark each claim confirmed, corroborated or inferred, and date the grade.
- Split notice-period from separated peopleSourcer, ongoing. Tag each person with an estimated end date and status, and replace estimates with dates candidates give you.
- Run the legal checkTalent lead, 1 to 5 days. Clear or hold each segment against the three legal triggers, in writing.
- Publish the dated outreach planRecruiter. Write segments with grade, counts at each step, and a send date per segment.
Here is the intake record I keep per filing. Paste it into the sheet before step 1 and fill it as you go; every blank is a step you have not done.
Company (operating): Legal entity on filing: State and portal: Notice type: full 639.7(e) / short 639.7(f) / unknown Sites listed: Per-site title rows: yes / no Notice date: Separation date(s): Total affected: Title rows (title :: count :: function): Siblings (state :: sites :: count :: notice date): Company framing (quote) :: grade: inferred until backed Team estimates (team :: count :: confirmed / corroborated / inferred :: date graded): Legal triggers hit: MSA-NDA-partner / exec non-solicit / in notice period / none Segments (name :: pool count :: verified count :: send date):
One record per filing; add a row under Siblings for each related notice you find.
For the Microsoft case the slate's first segment would read: PM and TPM, function confirmed by 591 filed positions, pool from 2,439 index profiles, sample showing 21 of 25 still at Microsoft and 11 of 25 in Washington, send date set from each person's verified end date. Every number in that line came from a step above, which is the point: a slate where you can trace each count back to its source.
The first message to a separated, confirmed-segment candidate should be specific about the role and not about their layoff.
Subject: [Role title] on [team], [location or remote] Hi [first name], I lead hiring for [team] at [company]. We are filling a [role title] role that looks close to the [function] work you did at [previous employer]. The short version: [one sentence on scope], [one sentence on level and team size], [location or remote terms]. If the timing is useful, I can share the full brief and a 20-minute slot this week. If it is not, no reply needed, and I am happy to check back later. [Name]
Use only for people whose end date you have confirmed has passed; keep the role details real and drop any line you cannot back up.
How this goes wrong: failure modes and false positives
Most errors in layoff sourcing come from reading a filing as more than it says, or reading a profile as current when it is not. Each failure mode below shows what it looks like, what it costs, and the check that catches it.
- Reading a short-form notice as "no titles exist." If the filer used 639.7(f), the titles exist on site but are not public. Check the state PDF first, and do not infer a team mix from the headcount alone.
- Taking the company's framing as attribution. Microsoft said the cuts were about removing managers; the filings showed 17% managers. Until title counts or named posts back a company statement up, it stays inferred.
- Treating a multi-site total as one team. Microsoft's notice covered Redmond and Bellevue; Google's mixed Kirkland, Seattle, Redmond and remote workers. Without per-site title rows, do not split counts by site.
- Missing cuts below the threshold. Pluralsight laid off 400+ people and filed no WARN notice because the cuts were spread across multiple locations. No tech-specific figure exists for how many cuts never produce a filing; all-industry estimates exist only in secondary sources. Check Layoffs.fyi and trade press as well as the portals.
- "Ex-company" keyword false positives. Boomerangs and headlines where "ex-" refers to an earlier employer. Verify the current-employer field, not the headline.
- Misreading the notice-date column. Some WARNTracker rows show the notice date equal to the layoff date, which can mean people were told earlier or got pay in lieu. Ask the candidate for their actual end date.
- Treating notice-period employees as free agents. They are still employees during the paid notice period. Contact can collide with confidentiality or no-poach terms, so they go through the legal check.
- Over-reading small samples. The index samples are 25 profiles each. Regional splits are directional only; widen the sample before you set a quota on them.
A filing tells you what was cut by title. It cannot tell you which team, and neither can the company's press release.
Keeping the slate current after the first send
A layoff slate decays as people sign elsewhere, sibling filings appear, and grades firm up or collapse, so re-check it on a fixed cadence rather than treating the first version as final. The ServiceNow case shows the timescale: a rumour took about 8 months to become a filing.
What to re-run, and why:
- Sibling filings. Re-search the other state portals for the same company over the 90 days around the original notice. New filings can add sites or confirm a team.
- Grades. When a named post or a later filing appears, upgrade the estimate and date the change. When press attribution contradicts the title rows, the title rows win.
- End dates. Move people from the notice-period segment to the separated segment as their dates pass, and send on the date set in the plan, not before.
- Portal rules. Thresholds, notice periods and what each portal publishes are set by state rules and portal design, both of which can change. Before relying on a state's format, re-read its WARN requirements page and open one recent public file to confirm the columns.
- The pool. Re-run the role, place and end-date query so new profiles show up, then re-verify current employer on anything you have not contacted. Refolk keeps the query in plain English, so re-running it is a paste, not a rebuild.
Before the first message goes out
- The filing's legal entity is matched to the operating company and every site maps to a unit or office.
- Sibling filings in other states and within the 90-day window are listed, or recorded as none found.
- Multi-site counts are kept whole unless the filing gives per-site title rows.
- Every team estimate carries a grade of confirmed, corroborated or inferred, with the date it was graded.
- The company's own framing is recorded and graded inferred unless title rows or named posts back it.
- Every person in the pool has a verified current employer, not a headline keyword match.
- Every person is tagged notice-period or separated, with an estimated end date.
- Each segment hit by an MSA, NDA, partner agreement, executive non-solicit or notice-period status has written legal clearance.
- Each segment on the slate shows its count at every step and a send date.
Questions practitioners ask
Do WARN notices list the job titles of affected employees?
A full state notice does. Under 20 CFR 639.7(e), the notice to the state must list the job titles of affected positions and the number of affected employees in each classification. The short alternative under 639.7(f) needs only the first separation date and a headcount, and the employer keeps the title detail on site. Whether titles are public depends on the state: California, Washington and Georgia publish them, and New York's dashboard display of titles is unverified.
When should I reach out to laid-off candidates?
Reach people who have already separated first, and treat people still in their paid notice period as a separate, legally checked segment. Notice periods can run weeks: Google's California notices gave 41 to 62 days. No published study compares response rates by timing, and nobody has published when recruiter message volume peaks after a filing, so date each segment from the separation schedule and the candidate's own stated end date.
How do I find out which teams were cut in a layoff?
Start from the job-title counts in the state filing, group them into functions, and compare each function's share of the cut with its share of the company's visible population. Then grade each team claim: confirmed if the filing states it, corroborated if press or named public posts agree, inferred otherwise. Microsoft's filing showed 17% managers despite a 'fewer managers' framing, which is why the company's own story starts as inferred.
Is it legal to recruit someone still in their notice period?
Often it is, but get it checked first. During a paid notice period the person is still an employee, and contact can run into confidentiality terms or a no-poach clause in an agreement between your company and theirs. An employee's severance non-solicit binds them, not you, but an MSA no-poach clause can bind your company. Some clauses explicitly carve out laid-off workers, so legal should read the actual text.
What if the company cut staff but never filed a notice?
Then the filing route gives you nothing, and you work from press and public posts instead. Cuts can fall below state thresholds or be spread across sites so no single site triggers a notice: Pluralsight laid off 400+ people without a WARN filing for that reason. Check Layoffs.fyi and trade press, and grade every team estimate from those sources as corroborated at best.
How do I find laid-off engineers from a specific company without false positives?
Filter on the current-employer field and the end date of the role, not on 'ex-company' headline keywords. In a 25-profile sample from Refolk's index, 9 US software engineer profiles with 'ex-Microsoft' in the headline still listed Microsoft as their current employer, because of boomerangs or because 'ex-' referred to an earlier employer. Confirm the end date with the candidate before you treat them as available.
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.