The Proven-Team Diaspora Teardown: Team Era to a Located, Ranked Slate
Rebuild one team's roster for a set date window, confirm each member with a dated artifact, find where they work now, and hand over a ranked slate with drop counts.
Key takeaways
- LinkedIn's documented tenure filters cover only the current job, so no past-company filter can tell you who was at an employer during a given window: you have to read and bound each profile's dates by hand.
- In Refolk's index, the headline keyword "payments infrastructure" returns 1,197 US profiles against 12 with exact payments engineering titles, and the keyword pool is led by CEOs and founders. That means neither search finds an engineering team.
- The median retroactive edit to a LinkedIn job entry lands 4.5 years after the job ended, which puts a team from about five years ago right in the window where entries get rewritten. Every member therefore needs a dated non-profile artifact.
- Public commits only show when they land on the default or gh-pages branch from a linked email, and private work is hidden by default, so a missing commit does not mean someone was off the team. Report a confirmation rate instead of dropping unconfirmed people without saying so.
- Among 12 US payments engineers in Refolk's index, the largest single employer holds 2 (17%), so three or more confirmed team members at one company is a regroup cluster worth turning into the next search.
- Run the off-limits check twice: once on the source company before you pull, and again after you resolve current employers, because an alumnus's new employer may be a client.
The request sounds simple: "get me people who were on that company's payments infrastructure team when it scaled, roughly five years ago." This teardown is for in-house recruiters, sourcers, talent leaders, and founders who have to turn a team and an era into a named roster, each person's current employer, and the two or three companies where the team regrouped. I carry one case through every cut, wrong turns included, so you can run your own case alongside it.
Why doesn't the past-company filter find a team?
A past-company filter returns anyone who ever listed the employer, with no tenure dates and no team attached. It is the starting pool for the roster, never the roster itself. Every guide that ranks for this job stops at "apply the past-company filter," and that is exactly where the real work starts.
Here is what the filter documents. LinkedIn's Sales Navigator help describes the Past company filter only as "Filter by leads' past companies." The company filter can be set to:
- Current
- Past
- Current or past
- Past not current
The only tenure filters cover the current job: "Years in current company" (which filters by the "range of years since leads joined their current companies") and "Years in current position." No documented filter sets tenure dates at a past employer. Third-party guides suggest Recruiter works the same way. They list Seniority, Years in Current Position, and Years in Current Company as Corporate-only filters, and none mentions a past-tenure filter.
The best public workaround is to put the years in as keywords next to the company filter. It is weak. A year keyword matches that year anywhere on the profile, including education and other jobs.
One thing is not established publicly: whether the company filter matches only the linked company page or also free-text employer names. You can test it on your own case. Run the filter on the company, then search the company name as a plain keyword with the filter off, and compare who appears in one result set but not the other.
The two shortcuts that look like they work
Before running the filter, most sourcers try a title search or a headline keyword. I ran both in Refolk's index for this case. They show why neither search finds a team.
| Query (Refolk's index) | Count | Largest single employer | Largest employer's share |
|---|---|---|---|
| Exact payments titles, US | 12 | Apple, 2 | 17% |
| Exact payments titles, UK | 2 | 1 each | 50% |
| Headline keyword "payments infrastructure", US | 1,197 | 1 each in a sample of 25 | 4% of sample |
The mechanism is simple. Engineers keep generic titles like "Software Engineer," while business people write the domain into their headlines. So the title search undercounts the team, and it only finds people who hold the title today, which is useless for alumni. The keyword search mostly catches the wrong people: no engineer title appears in its top 10.
Two shortcuts, neither one the team
- 1,197Headline keyword "payments infrastructure", US
Top sampled titles are CEO and Founder
- 12Exact payments engineering titles, US
Current holders only, not alumni
- 2Largest single employer in the title pool
Apple; Square, Google, Airbnb, Brex have 1 each
The worked case: what does the request become once it's written down?
The request turns into a written spec with four parts: a named team, a dated window, a full list of the company's entity names with effective dates, and a cleared off-limits check on the source company. If any of the four is missing, the roster will drift.
For this teardown, the spoken request was "Square's payments people, back when it scaled." After a 30-minute call with the hiring manager, it became:
- Team: payments infrastructure, which the manager also called the payments platform team.
- Window: 2019 to 2021. "When it scaled" was the manager's memory, so I asked which launches or incidents bracketed it and wrote down the years they agreed to.
- Entities: Square, plus Block, which some alumni list instead. Former names, predecessors, and subsidiaries get added in Step 2.
- Off-limits: I checked the source company against current no-poach and client agreements before pulling anyone.
Resolving the aliases
Nobody has published what share of profiles list a parent company, an acquired predecessor, or a renamed entity instead of the unit the person actually worked in. Assume the share is large enough to matter and resolve the aliases before you pull.
For UK companies, the process is documented. On the Companies House register, the company number stays the same for life, even through a name change, and the register shows previous names with the date each change was registered. Companies House staff have confirmed "we store all of the former names." Some records carry a long history: company 00201671, for instance, has 10 former names. A name change is filed on form NM01, so each rename leaves a dated filing.
I did not verify an equivalent procedure for US merger or subsidiary records, which matters for a US company like Square. In that case, build each alias's date range from public records, write down the source next to each date, and flag any range you could not pin down.
The entity layers one team can sit under
- Parent or group nameBlock, the other name some alumni list
- Operating brandSquare, the name most alumni list
- Former names and predecessorsEach alias with an effective-date range from the public record
- Subsidiaries and acquired unitsStaff who sat on the team under another entity name
- The team in the windowPayments infrastructure, 2019 to 2021
TEAM: Payments infrastructure (also called: payments platform) SOURCE COMPANY: Square ENTITY ALIASES (name :: effective from :: effective to :: source): Square :: ____ :: ____ :: ____ Block :: ____ :: ____ :: ____ [former name / predecessor / subsidiary] :: ____ :: ____ :: ____ ERA WINDOW: 2019 to 2021 (anchored to: ____) OVERLAP RULE: any month of tenure inside the window counts; record months of overlap FUNCTIONS IN SCOPE: software engineering, infrastructure/SRE, engineering management of this team FUNCTIONS OUT: sales, support, account management, non-engineering roles ARTIFACTS ACCEPTED: public org commit, patent with assignee match, talk, paper, engineering blog byline OFF-LIMITS CHECK (source company): cleared by ____ on ____ DELIVERABLE: ranked slate + confirmation rate + drop count per step + regroup employers
Fill this in on the call with the hiring manager and get them to sign off on the window before you pull anyone.
The procedure: from team and era to a ranked slate
The procedure has eight steps, and the count after each cut goes in a log. The log is the deliverable as much as the slate is. No source documents how long these steps take, so the times in the walkthrough below are my unverified estimates.
Team-era roster to ranked slate
- Scope the team, window, and entitiesPin down the team name, era window, and entity list with the hiring manager, then check the source company against off-limits agreements. Done when the window is written and the source is cleared.
- Resolve every alias with effective datesPull former names with registration dates, plus predecessors and subsidiaries. Done when every entity name has an effective-date range.
- Run the raw past-company pullRun Past not current across every alias and record the raw count. Split by alias or location if a search nears the reported 1,000-result cap.
- Bound every profile to the eraRead each profile's dates for the alias entity and keep only tenures that overlap the window. Done when the overlapping count is logged.
- Cut by functionDrop sales, support, and other non-engineering roles. Done when the post-cut count and the number dropped are logged.
- Confirm membership with a dated artifactFind at least one dated non-profile artifact per person: public org commit, patent, talk, or paper. Done when the confirmation rate is stated and unconfirmed people are labelled, not deleted.
- Resolve and re-verify current employersConfirm each current employer with a second dated signal and rerun the off-limits check. Done when every person has a current employer and a verification date.
- Cluster by current employer and rankCount confirmed alumni by current employer, flag three or more at one company, and rank. Done when the slate shows every step's drop count and notice is ready for first contact.
Not everyone agrees on the order. Some sourcers cut by function (Step 5) before bounding the era (Step 4), because a title is quicker to read than a set of dates. Nobody has published a comparison of the two orders. I bound the era first, because people who joined after the era are the biggest source of inflation in this job, and a title read before the dates misleads you about which role was held in the window.
Teardown: what happens between the raw pull and the era-bounded roster?
Steps 3 to 5 shrink the pull to the people whose tenure at any alias overlaps the window, in an engineering function. Most of the list inflation comes from three wrong turns made here.
Wrong turn 1: pulling on the brand name alone
My first pull used Square only. It missed everyone who lists Block, and it would have missed anyone who lists a predecessor or subsidiary. The fix is to run one Past not current pull per alias, then deduplicate. A third-party guide reports a 1,000-result cap on every LinkedIn search whatever the account type, which is unverified with LinkedIn. If any alias's pull comes back at or near that number, assume it was truncated and split it by location or by function keyword.
Wrong turn 2: trusting year keywords to do the era cut
Adding "2019" and "2020" as keywords felt like a shortcut. It kept people whose only 2019 was a graduation date, and it dropped people whose job entries showed no years in the text. You have to read the dates. At roughly 1 to 2 minutes per profile, this is the step people skip, and it is the one that makes the slate defensible.
Two types of record leave the list here:
- Joined after the era. The person was at Square, but their first month there falls after the window closes. This is the most common false positive in this job.
- Left before the era. Their tenure ended before the window opened. They may know the founding engineers, but they were not on the team the manager means.
Log both drops separately. The manager will ask how many people you cut for timing.
Wrong turn 3: assuming the employer filter means engineering
The past-company filter matches the employer, not the function. It catches sales, support, account management, and everyone else who listed the company. Drop them in Step 5 and log the count.
Cutting early matters for more than effort. Under UK GDPR Article 14, the notice clock starts when you collect the data, so a large raw pull that sits unworked carries notice liability. Every record you discard at Steps 4 and 5 is one you are no longer holding.
Which artifacts confirm team membership?
A dated record outside the profile confirms membership: a commit in the company's public org repos, a patent naming the company as assignee, a talk, or a paper. The profile entry alone does not, because the window this manager asked about sits inside the period when people most often rewrite old entries.
The same research shows how often that happens. Retroactive edits appear on 38.1% of profiles of users under 30, and the language of those edits drifts: AI wording "rises sharply in retroactive edits" after the period it describes. A job from that window now labelled with fashionable wording tells you about the person's current positioning, not what they did then.
| Artifact | Date carried | Documented coverage limit |
|---|---|---|
| Public commits | Commit date | Default or gh-pages branch only; linked email only; private work hidden |
| Patents | Filing date plus assignee | Disambiguated quarterly; raw assignee fields not in the API |
| Profile job entry | Self-reported | 19.7% retro-edited, median 4.5 years after the job ended |
| Talks, blog bylines | Event or post date | Not established publicly |
| Paper co-authorship | Publication date | Not established publicly |
Commits: what they prove and how they lie
A commit to the company's public org repo, dated inside the window, proves the person wrote code under that organisation in that window. GitHub's documented rules limit what you can see. Commits only count if they are made "in the default branch or the gh-pages branch" with an email linked to the account. Private activity "doesn't appear by default," and commits made with generic emails "will not show up in your contribution graph." New contributions can take up to 24 hours to appear.
So read the git log of the company's public org repos directly instead of a person's contribution graph. The log shows author emails and dates that the graph hides. Payments code is usually private, so most infrastructure engineers will have nothing here, and an empty result means nothing.
Patents: what they prove and how they lie
A patent naming the person as inventor and the company (or an alias) as assignee, with a filing date in the window, is strong evidence. PatentsView "disambiguates patent inventors, assignees, lawyers, and locations" and updates "on a quarterly basis." Raw assignee fields are "not currently being offered by the API," so you are relying on its disambiguated assignee, which is one more reason to search every alias from Step 2. Patents lie through name collisions: a different inventor with the same name. Match on assignee, filing date, and co-inventors, which are the signals PatentsView itself uses.
Talks, blog bylines, and papers
A conference talk or engineering-blog post with the person's name and the company in the window is useful, but nobody has published how complete these sources are. Use them as confirmation when you find them, and do not read their absence as anything.
Joining these sources is the slow part of the job: one person's commits, patents, and talks sit in three places under three spellings of their name. Refolk runs that join across GitHub, LinkedIn, and the open web from a single plain-English request, so the dated artifacts arrive attached to the person instead of in separate tabs.
When Step 6 is done, state the confirmation rate as confirmed people divided by people remaining after the function cut. I don't have a published benchmark for what a good rate looks like, so report yours as a fact about this case, not as a grade.
Where does the team work now, and where did it regroup?
You resolve each person's current employer with a second dated signal, then count confirmed alumni by employer. Any company holding three or more confirmed members is a regroup cluster, and that list of companies becomes your next search. The profile's current employer is a claim, not a fact. It goes stale, and the lag is poorly measured.
| Clock | Value | Status |
|---|---|---|
| Article 14 notice | One month, or at first contact | Statutory |
| Profile edits before a move | Rise about 6 months before | Measured (NBER 35546) |
| Vendor data lag | 0 to 120+ days | Modelled, not benchmarked |
| Profile self-update lag | 27 to 45 days | Unverified secondary claim |
For the second signal, use anything dated that ties the person to the new employer: recent commits to that company's public org, a talk under the new affiliation, a patent filing with the new assignee. Write the verification date next to each employer. Then run the off-limits check again, this time against the new employers. A cleared source company tells you nothing about where the alumni went.
Why clusters are real
Published labor research backs the idea that teams re-form. In Austrian administrative data, 21% of displaced workers find a new job at a firm connected to their former workplace, and workers with a link to a former coworker are twice as likely to be hired by the connected firm. Danish data from 1980 to 2005 show "a strong tendency of workers to follow their former coworkers into the same establishments." Referrers are drawn disproportionately from the referred workers' hiring cohorts, and workers' networks are front-loaded with people they met early in their careers. A team that scaled together is exactly that kind of cohort.
Nobody has published a threshold for when three or more people at one employer is a signal rather than chance. My working line comes from Refolk's index: among 12 US payments engineers, the largest single employer holds 2, or 17%. Three or more confirmed members of one specific team at one employer is above that national spread. Treat it as derived reasoning, and say so on the slate.
Freshly edited entries as a timing signal
The edit research also helps with ranking. The probability of a profile edit "begins to rise about six months before the transition," and workers are 66% more likely to change employers in a period when they update. If someone has just rewritten their entry for the era, that is a reason to call them sooner. It is also a reason to confirm the entry with an artifact.
Who to call first
The cluster list also sets up the next search. Once you know which companies the team regrouped at, the follow-up query takes the form of "Former Braintree or PayPal payments engineers who now work at Brex, Ramp or Plaid": known source, known destinations. Refolk accepts that kind of follow-up as written, which is how one teardown feeds the next.
How does this go wrong?
Most failures in this job inflate the list with people who were never on the team, and a few quietly drop people who were. Each failure has a specific check, and the slate is only as good as the checks you actually ran.
| Failure | What the false result looks like | Check |
|---|---|---|
| Past-company filter treated as the team | Anyone who ever listed the employer, whenever they were there | Bound every profile's dates by hand (Step 4) |
| Headline-keyword pull | Founders and CEOs who mention payments infrastructure | Require an engineering title on the era-window entry |
| Retro-edited era titles | A role from the window relabelled years later in fashionable wording | Require a dated non-profile artifact |
| No commits read as "not on the team" | Infrastructure engineers dropped because their work is private | Read public org git logs; label, do not drop |
| Patent name collisions | A different inventor with the same name | Match assignee, filing date, co-inventors |
| Stale current employer | Profile still shows the old employer | Confirm with a second dated signal |
| Off-limits check run once | Source cleared, but a new employer is a client | Rerun the check at Step 7 |
The two failures that cost the most in practice point in opposite directions. Joiners from after the era inflate the roster, and the hiring manager notices first because they know who was there. Unconfirmed engineers dropped without a word shrink it, and nobody notices at all, which is worse. Engineers who did the most infrastructure work are the most likely to have only private commits.
A slate that silently drops unconfirmed engineers is hiding its own error rate.
Handing over the slate and keeping it current
The handover is a ranked slate plus a step log that shows the drop count at every cut, the confirmation rate, and the regroup employers. The manager should be able to see why each name is on the slate and how many people did not make it.
TEAM / WINDOW: Payments infrastructure, Square (incl. Block and aliases), 2019 to 2021 STEP LOG Raw past-company pull, all aliases, deduplicated: ____ After era bounding: ____ (dropped: joined after window ____, left before window ____) After function cut: ____ (dropped: ____) Artifact-confirmed: ____ of ____ (confirmation rate ____%; profile-only ____) Current employer verified: ____ (unresolved ____; excluded off-limits ____) REGROUP EMPLOYERS (3+ confirmed members; derived threshold, not a published one) ____ : ____ members ____ : ____ members RANKED SLATE Rank | Name | Era role and dates | Artifact (type, date) | Current employer | Verified on | Cluster | Notes 1 | ____ | ____ | ____ | ____ | ____ | ____ | ____
Fill every count from your log; leave none blank, and write "not checked" rather than guessing.
Before first contact, sort out notice. Under UK GDPR Article 14(3), notice is due within one month, or at the time of the first communication at the latest if you use the data to contact the person, and also when the data is first disclosed to a client. The ICO's draft guidance says recruiters are likely to be controllers in most circumstances. Its audit of AI recruitment tools described a provider that "sourced candidate profiles from public job networking sites and data vendors" and recommended active notice within one month. It also turned down reliance on the disproportionate effort exemption. The mechanics are in the first-contact notice guide.
Before the slate leaves your hands
- The era window is written down and signed off by the hiring manager
- Every alias has an effective-date range with a source next to it
- Each alias was pulled separately and no pull sat at the reported 1,000-result cap
- Every person's tenure was read by hand and overlaps the window
- Drops for joining after the era, leaving before it, and function are logged separately
- Every person has a dated artifact or the label profile-only
- The confirmation rate is stated on the slate
- Patent matches agree on assignee, filing date, and co-inventors
- Every current employer has a second signal and a verification date
- The off-limits check was run on the source company and again on current employers
- Regroup employers are listed with member counts and marked as a derived threshold
- Article 14 notice is drafted to go out at first contact
To keep the slate current, re-verify current employers before each outreach wave instead of trusting the date on the slate. The vendor-lag model runs to 120+ days, and nobody has a benchmarked figure. Rerun the patent pass after PatentsView's quarterly update, because new filings and disambiguation fixes appear there first. Treat every regroup employer as a new search: run its alumni-in-cluster list through the same eight steps with that company as the source, and the team you started with will lead you to the next one.
Questions practitioners ask
Can I filter LinkedIn by the dates someone worked at a past company?
No documented filter does that. Sales Navigator's tenure filters, Years in current company and Years in current position, only cover the current job, and third-party guides to Recruiter mention no past-tenure filter either. The best public workaround is to add years as keywords next to the company filter, but that matches the years anywhere on the profile. The reliable method is to pull everyone with the past employer and read each profile's dates by hand.
How do I narrow a past-company search down to one team?
Go in three cuts. First, bound each profile to the era window by reading the dates for that employer. Then drop non-engineering functions such as sales and support. Finally, confirm team membership with a dated artifact: a commit in the company's public org repos, a patent naming the company as assignee, or a talk or paper. Write down the count after each cut so the manager can see how the list shrank.
What if a team member has no public commits?
Keep them on the slate as unconfirmed, not dropped. GitHub only shows commits made to the default or gh-pages branch from an email linked to the account, and private work is hidden by default. Payments code is usually private. Read the git logs of the company's public org repos rather than the contribution graph, then look at patents and talks. If nothing turns up, label the person profile-only and count them in the confirmation rate.
How many alumni at one company counts as a regroup cluster?
No published threshold exists, so treat three or more confirmed team members at one current employer as the working line. That line is derived reasoning. In Refolk's index, the largest single employer among 12 US payments engineers holds only 2, and published labor research shows a link to a former coworker doubles the odds of being hired by that firm. Three is above both the national spread and chance.
Do I need to notify people I source from public profiles?
Under UK GDPR Article 14, yes. Notice is due within one month of collecting the data, or at the first contact at the latest if you use the data to contact the person. The ICO's draft guidance says recruiters are likely to be controllers in most circumstances, and its audit report turned down reliance on the disproportionate effort exemption. Cutting the raw pull early reduces how much data carries that obligation.
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.