The Stealth Founder Signal Reference: Reading One Person's Pre-Launch Footprint

You are looking at one person's public footprint and need to judge whether they are quietly forming a company, and how much to trust each clue in front of you. This reference is for early-stage investors, platform and talent partners, and angels who source deals before they are announced. It treats each pre-formation signal on its own row so you can jump to the one signal in front of you, read what it proves, how it misleads, and how fast it decays, and rank the person accordingly.

This is not a workflow for building a tracking system. It is a lookup document for reading a single individual. The core finding to carry through every row: the earliest signals are the least certain, so a cluster of co-occurring signals from one person beats any single flag, and the strongest single signal is an absence rather than a presence.

What signals exist, and are they about the person or the company?

There are seven documented pre-formation signals, and the first thing to know about any one of them is whether it describes the person or the footprint of a company they may be forming. That split governs both lead time and reliability.

People-level signals describe the individual directly: a title change, a departure, GitHub activity, and cofounder-search behaviour. Footprint signals describe a company that is starting to exist on paper: a new domain, an incorporation or trade-registry filing, and a patent. Practitioner and vendor write-ups converge on this same set.

The reason the split matters: employment and network signals come first in time but are ambiguous, while footprint signals lag by weeks but are structured and harder to fake. A network signal adds something an employment signal cannot - that the person is recruiting people to join them, which is pre-product evidence of company formation. Taken together, employment plus network signals distinguish a stealth founder departure from any other senior departure.

How fast does each signal arrive, and how fast does it decay?

Lead time and reliability trade off against each other. The earliest signals are the least certain; the most certain arrive last, sometimes after the company already exists. No source publishes a dated distribution, so the timing below is directional.

Two structural facts anchor the slow end of that table. A provisional patent gives the filer 12 months to file a non-provisional claiming priority to it, and an application becomes publicly available 18 months after filing unless a non-publication request was filed. So a patent you can see almost always reflects a company that already launched. It is confirmation, never early warning.

Decay works in the other direction for the fast signals. A departure with no public landing is a strong flag in its first weeks; if six months pass and the person still has no company on paper, the same record now argues for a sabbatical or a long job search rather than a founding. Down-weight every signal as it ages, and re-run the read on a schedule rather than trusting a snapshot.

Reading each signal row by row: what it proves, how it lies

Each signal below stands alone. Jump to the one in front of you. Every row states what the signal proves when honest and what it looks like when it misleads.

The stealth or building-something-new title

What it proves: the person wants to be seen as pre-formation. It is the easiest signal to query and therefore the easiest to stage.

How it lies: some founders keep a visible stealth title as inbound bait. More VC inbound happens in stealth mode, because for VCs scouting early-stage deals, searching for founders with stealth in a LinkedIn title is trivial. So the title population includes people farming attention without an incorporated company, alongside consultants, freelancers, and people between jobs. Filters are needed to narrow to people building real companies.

The departure with no public landing

What it proves: the person left a named high-density employer and has attached no new one. This is the strongest single signal because a missing next-employer is costly to fake, while a title is free to author. A repeat founder or strong operator's departure tends to precede their next startup by weeks or months.

How it lies: a removed employer can mean a layoff, a sabbatical, or a quiet job search. The check is the network signal - are they recruiting? That is what distinguishes a stealth founder departure from any other senior departure.

The new GitHub organization repo

What it proves, when corroborated: early product development. A developer who recently left a senior role and starts pushing to a new organization-level repository is worth flagging, especially if the repo is private but visible through contribution graphs.

How it lies: most new repos are side projects, not ventures. Require a README that describes a product, multiple contributors joining within weeks, consistent commit velocity, and commercially-suggestive repo naming before you treat a repo as founding activity.

The cofounder-search or founding-hire post

What it proves: intent to assemble a team. When someone posts for a Founding Engineer, First Designer, or Early GTM Hire, they are likely building.

How it lies: a cofounder-matching profile can be exploratory. When YC's platform ran at scale it showed a 25% invite-acceptance rate, so most searchers never convert. Corroborate with a departure or footprint signal.

The fresh domain

What it proves: something was named. Weakly.

How it lies: heavily. Legitimate startups launch, enterprises create sub-brands, and new products get dedicated landing pages, so a domain under 30 days old may be entirely benign. WHOIS privacy on many TLDs masks the registrant name, organisation, and email. Use the creation date as a weak timing signal only, and tie the domain to the person, not to a matching name.

The incorporation or registry filing

What it proves: a company legally exists. This is a strong structured signal, but it lags departure by weeks and only appears once the person has committed to paper.

The patent filing

What it proves: confirmation of a technical company, already formed. Use the filing or priority date, not the publication date. A law-firm figure claims startups with early patent filings are 6.4 times more likely to secure venture capital, though that claim's methodology is not shown.

The strongest signal a founder can give is the one they cannot cheaply fake: a departure with nowhere listed to land.

Why the strongest read is a cluster, not a single flag

The single strongest read is a departure from a high-density employer combined with no public landing at a new one, and the strongest read overall is several signals from the same person inside one short window. When multiple signals from the same individual appear in a short time window, that combination is a high-confidence detection of founding activity.

This is a direct consequence of the lead-time trade-off. Because the earliest signal is always the least certain, no single early flag is trustworthy on its own. Clustering fixes that: a stealth title plus a departure plus a founding-engineer post plus a fresh domain, all in six weeks, is a different object than any one of those alone. Rank clustered, time-correlated signals from one person far above any single signal, however strong that single signal looks.

Some markets make clustering harder because the raw pool is thin. In Refolk's index, the US carries 315 stealth-titled people against 44 in the UK - a 7.2x gap. A UK-focused reader working the same title filters sees a far thinner pipeline, so footprint and network signals carry more of the weight there.

The phrasing you filter on also changes the population. In the US, a bare stealth title returns 315 people, while a Founder title paired with building something new returns 209 - about 66% as many. The most common employer string in that sample was Stealth Startup.

Running these title-and-departure filters by hand across public LinkedIn and the open web is where most of the time goes. Describing the person you want in plain English and getting the matched set back removes that friction.

The read: seven steps from profile to ranked likelihood

Work one person at a time. The order below puts the early, ambiguous signals first and lets the later structured records confirm or kill them. Sources disagree on one point: some vendor checklists sweep domains and registry filings first because they are queryable at scale, while the fuller read places footprint after employment and network. This procedure follows the fuller read.

The output of the read is a rank, not a yes or no. A person with a departure-plus-absence and a founding-engineer post inside a month sits near the top. A person with only a stealth title sits in a holding pen until a second signal lands or the flag decays.

How this read goes wrong: the false positives

Most bad calls trace to one of seven failure modes. Each has a specific tell and a specific check. This is the part of the reference worth the most attention, because ranking a false positive costs a fund meetings and credibility.

1. Stealth title as inbound bait. An operator keeps a stealth title to farm VC inbound without an incorporated company. Check: require a footprint or hiring signal before ranking up.
2. Blank employer read as founding when it is a job search. A removed employer can mean a layoff or sabbatical. Check: look for the network signal, the recruiting behaviour that separates a founder departure from any senior departure.
3. Fresh domain treated as a company. The newly-registered signal alone generates false positives, because legitimate startups, product launches, and marketing campaigns all create domains. Check: tie the domain to the person, not just to a matching name.
4. WHOIS privacy read as identity. On many TLDs the registrant name, organisation, and email are masked. Check: use the creation date only as a weak timing signal; never infer identity from a masked record.
5. Patent read as early warning. The 18-month publication lag means a visible patent may reflect a company already launched. Check: use filing and priority dates, treat as confirmation.
6. GitHub side project mistaken for a venture. Check: require a product README, multiple contributors within weeks, consistent commit velocity, and commercially-suggestive naming.
7. Cofounder-matching profile read as a committed founder. A matching profile can be exploratory given the 25% acceptance rate at scale. Check: corroborate with a departure or footprint signal.

Person:
Current title: [stealth / founder / building something new / other / blank]
Employer field: [named / removed / blank]
Departure: [employer name] left [date] -> next employer: [named / none]
Network signal: [founding-hire post? y/n] [ex-colleagues moving? y/n]
Footprint: [domain age] [incorporation? y/n] [patent filing date, not publication]
Technical: [new GitHub org repo? y/n] [product README? y/n] [contributors joining? y/n]
Social intent: [cofounder-matching profile? y/n] [cofounder-wanted post? y/n]
Signals in the last 8 weeks: [count]
Rank: [top / hold / discard]  Re-check on: [date]

Keeping the read current

A pre-formation read is a snapshot of a moving target, so schedule a re-run rather than trusting one pass. The signals decay at different rates: a departure loses force after a few months without a company, a fresh domain ages out of the under-30-day window, and a patent that publishes turns from absent to confirmatory 18 months after it was filed. Re-run the read on your holding-pen names at least monthly.

Two mechanisms move faster than your snapshot and are worth watching directly. First, the stealth population itself is growing - the number of people working for startups in stealth has more than tripled in two years per LinkedIn data - so title filters return more people and more noise over time, which raises the value of corroborating signals. Second, cofounder-matching activity is now large: three months after launch, YC's platform had 16,000 profiles and 33,000 matches, so a matching profile is common and weakly predictive on its own.

When you keep this reference open beside a profile, the discipline is simple: read the one signal in front of you for what it proves and how it lies, refuse to rank on a single flag, weight the cluster, and set a date to look again. The people worth your time will show an absence you can trust before they show a title you cannot.