if you are evaluating tools for finding founders before the company exists, the useful first cut is not price or coverage. it is which of two families the signal belongs to.
family one: artifacts
an artifact is something the founder filed or published. the source reads a record.
- trade registry filings and incorporations
- domain registrations, whois and dns records, certificate transparency logs
- github activity: a new org, a private repo going public, a commit burst
- patents and research papers
- grant awards and commercialisation programmes
- job posts, especially a founding-engineer role at an unnamed company
- self-declared linkedin titles: "stealth", "building something new"
what an artifact proves: an entity now exists, or a person has taken an action consistent with starting one.
properties: high reliability per signal, because a filing is a fact. low selectivity, because most filings are not interesting. and critically, artifacts are durable and public, which means anyone running the same crawler finds them at the same time, and they can be crawled retroactively.
evertrace is the clearest example of an artifact-first product. launch gravity is a single-artifact product, reading one linkedin field at volume.
family two: behaviour
behaviour is who started paying attention, and when. the source reads a change, not a record.
principally: which venture investors began following a person or a young account, and how many of them converged in what window.
what behaviour proves: nothing about whether an entity exists. it indicates that people whose job is making this judgement have already made it.
properties: noisier per event, much more selective, and it carries an implicit quality filter that no artifact has. a domain registration does not know whether the person is any good. thirty-three investors following someone is thirty-three assessments.
the asymmetry that decides the evaluation
artifacts can be bought later. behaviour cannot.
a registry filing from march is still there in september. a patent, a domain record, a github commit, all still there. if you start crawling tomorrow you get the back catalogue.
a follow-graph diff is not a record, it is a change. looking at someone's followers today tells you who follows them; it does not tell you when each one arrived. that information exists only if something was recording on the day it happened. nobody can reconstruct who started following a founder last march. not us, not a competitor with more money, not the platform.
which cuts both ways, and we should say so: it means our own history only goes back as far as we were watching, and there are founders we missed because we were not tracking the right investor yet.
what each family is actually good at
| question | artifact source | behavioural source |
|---|---|---|
| does this company legally exist | yes | no |
| is anyone credible interested | no | yes |
| can I get the back catalogue | yes | no |
| will my competitor have the same data | yes | only if they were also recording |
| does it work without the founder self-declaring | mostly yes | yes |
| how much noise per signal | low | higher |
| how much filtering is needed | a lot | a lot |
the practical answer
run one of each, and require them to agree.
an artifact plus behaviour is a much stronger position than either alone: a domain registered by someone three tracked investors just started following, whose career says they can build it, is a very different row from a domain registered by nobody in particular.
that is also the bar we use internally before a person is shown to a customer. of 2,549 people flagged, 896 clear it. the 64% that do not are the ones where one signal fired and nothing corroborated it.
the sources, named
- linkedin and proxycurl: artifact, self-declared titles. free, real, self-selecting ceiling.
- evertrace: artifact, broad. registries, domains, github, patents, grants.
- launch gravity: artifact, single-source and high volume, with founder contact details.
- harmonic: artifact, company-first, index scale. starts later by design.
- frontrun: behavioural, plus a judged career check. 2,900+ tracked investors, dated flags, from $49/mo.
longer comparison including where each one wins at best tools to find founders building in stealth.
related
open source deal flow
companies and founders surfaced on frontrun. curated, every week.
