Sourcing Academic Founders Before They Leave the Lab

This is a sourcing procedure for early-stage investors, platform and talent partners, and angels who want to reach academic founders before they incorporate or raise. It turns a target field or set of labs into a ranked, contact-ready watchlist of pre-incorporation researchers, each tagged with the public signal that flagged them and a defensible window for first contact. It works the pre-departure window using signals anyone can read: paper and code releases, patent filings, invention disclosures, and grant and I-Corps cohorts.

Most spinout and SBIR guides start after someone has already left their job or won an award. By then the round is forming and the term sheet is a scramble. This one works earlier, when the person is still a named author on a paper and the only public trace of intent is a repository, a provisional filing, or a seat on an I-Corps team. The hard part is not finding researchers. It is resolving which named author is the engineer-founder versus the advisor who will stay in the lab, and knowing when the clock starts.

What signals mark a founder before incorporation, and in what order

The documented formation path starts with an invention disclosure to the technology transfer office, filed before any public talk, poster, or paper because public disclosure can trigger statutory bars to patentability. After the disclosure come customer discovery through I-Corps, an SBIR or STTR Phase I award, incorporation plus a license, and then a seed round. That is the textbook sequence, and it is a poor model for sourcing because the stages overlap.

Real technology transfer offices rarely run these stages as a strict, one-directional funnel. Equity negotiation and initial funding conversations often overlap, and a disclosure can sit for months awaiting a publication decision or a co-inventor's sign-off. No single standard formation duration is publicly established. For sourcing, that is good news: it means several independent signals exist at once, and you do not have to catch a person at one gate.

The signals you can actually read, and what each one proves:

Read the path as a set of overlapping windows, not a line. A person can have a repo and a provisional on file long before any I-Corps seat, and a seed conversation can open before the disclosure is resolved. The job is to catch them while at least one signal is public and the priced round is not.

Why code release stopped being a founder signal on its own

Code release is now too common to mean anything by itself. The discriminating signal is who committed the code, not whether code exists at all. At the top machine learning venues, shipping a repository is majority practice by camera-ready, so the mere presence of a GitHub link tells you almost nothing about founder intent.

The share of NeurIPS papers containing open-source code rose from 27.6 percent in 2016 to more than 60 percent from 2019 onward. When a signal describes the majority of a field, it cannot rank anyone. So treat code as a gate, not a score: use it to confirm the work is real and reproducible, then read authorship and committership to find the person.

The founder-resolution convention is simple and worth stating plainly: the first or corresponding author usually did the work, and the last author is usually the principal investigator who stays in the lab. Match the repository's top committer against the author list, and the engineer-founder tends to fall out. This is a widely understood reading, not a published, validated method, so treat it as inference you verify per name rather than a rule you trust blind.

How to tell the operator from the advisor

The operator-versus-advisor read is the whole game, because an advisor who stays part-time is not a fundable founder. Read three public records together and let them agree before you commit a name to the top of the list. No validated quantitative scoring model is publicly established, so what follows is a weighting you apply and sanity-check, not a formula you can cite.

The three reads, and what each proves when it holds and when it lies:

• Code committership versus last-author position. A first or corresponding author who is also the top committer is doing the engineering. It lies when the committer is a grad student who intends to stay; pair it with an entrepreneurial signal before trusting it.
• I-Corps role. Teams run as three people: a technical lead, an entrepreneurial lead, and a mentor. The entrepreneurial lead is the operator signal. It lies rarely, which is why rosters are so useful, but the mentor and PI on the same team are not founders.
• Licensing intent. Pursuing an option-to-license reads as operator; pure licensing with no company reads as advisor. This is harder to observe publicly and often only surfaces in conversation.

Equity itself is a forward-looking signal, not a reward for the past. The biggest mistake scientific teams make is thinking the purpose of allocating equity is to reward past contributions, when it is mainly to anticipate future ones. When you read a cap table later in diligence, a researcher carrying founder-level equity is being bet on to build; one carrying 10 percent or less is being thanked for past work. A VC view argues academic founders who stay on as advisors should hold 10 percent or less, precisely because it is not appropriate to disproportionately reward someone for past contribution.

I-Corps rosters deserve special weight because they are a published operator-tagging list. The program's three-role structure names an entrepreneurial lead, and across the program more than 2,500 teams created nearly 1,400 startups, a roughly coin-flip conversion. That gives you a pre-incorporation shortlist with a known base rate, which is rare in sourcing.

Where to read these signals without a vendor

Four public sources carry the signals, and each proves a distinct thing, so you query them in sequence rather than treating them as interchangeable. Patents prove invention plus assignee, which answers who owns it. SBIR proves non-dilutive de-risking. I-Corps proves customer-discovery intent. Spinout deal-terms records prove the institution's equity appetite before you ever meet the team.

A note on patent data: historically the inventor and assignee API allowed around 45 queries per minute, which is plenty for a target field but not for a blind national sweep, so scope your field tightly first. Under Bayh-Dole, any IP resulting from federally funded award work must be registered in the iEdison database, and once title is elected a patent application must be filed within one year. Those registration and filing deadlines create dated records you can anchor an outreach window to.

The friction here is identity resolution: the same person appears as an author on one source, an inventor on another, and a team member on a third, under slightly different names and affiliations. Pinning that one human across sources is where most of the analyst hours go, and where a plain-language people search earns its place in the workflow.

With Refolk you describe the person in plain English and get back resolved profiles across the public web, which collapses the cross-source matching that otherwise dominates steps three and four. I built it so that a query like the one above returns people, not a list of papers you then have to de-duplicate by hand.

The procedure: from a target field to a ranked watchlist

Run the following seven steps in order, with the caveat that the formation stages themselves overlap, so your ranking must not assume a strict funnel. The time estimates assume one analyst and one investor working a single field; budget roughly eight to twelve working days from a cold start to a contact-ready list.

When to reach out: the outreach clock

A defensible outreach window is roughly 6 to 18 months before a priced round, and you anchor it on whichever milestone you can date. Two numbers set the bounds. The SBIR Phase I to Phase II gap runs 6 to 12 months, and it is the most common reason promising Phase I awardees never reach Phase II. The average UK spinout takes 12 to 18 months to launch from the point a founder decides to commercialise.

The SBIR gap is the sharpest clock because it has a mechanism behind it. During that 6 to 12 month gap the founder needs non-dilutive bridge capital, which is exactly the moment an investor can enter before a priced round. On the far side, YC notes founders typically need to work full-time on the company for a year or more before it is ready for a multi-million dollar VC round, so a person who has not yet gone full-time is early for a priced conversation but not for a relationship.

Data lag is your edge, not a nuisance. SBIR and grant databases often lag 12 to 18 months before details update, so anyone waiting for the award feed to refresh is months behind. A reader who resolves founders from papers and patents is ahead of that feed by design. Treat the absence of an award as unknown, not as a negative, and date your window from the signal you can actually read.

Data lag is the edge. Resolve founders from papers and patents and you are months ahead of the award feed.

How this goes wrong

The failure modes here are specific and they repeat. Each one has a false positive and a check that catches it. Work them before you spend investor time on outreach, because a padded watchlist costs more than a short one.

• Code presence misread as intent. A repo exists but is maintained by a grad student who will not leave. Check that the committer is the first or corresponding author and that an entrepreneurial signal exists, since majority code release makes presence alone weak.
• Last author mistaken for operator. The last author is usually the PI who stays. Check committer identity against the author list before ranking anyone.
• Patent assignee confusion. Keyword matching on assignee pulls unrelated entities, the classic case being "Apple" against "Appleton Papers." Cross-verify the assignee entity before trusting a hit.
• Stale grant data. Databases lag 12 to 18 months, so "no award" may be false. Check filing dates and treat absence as unknown.
• University equity kills the deal after identification. A high-equity institution, UK average 19.8 percent and some deals above 60 percent, can make a strong team unfundable. Check the institution's policy on a spinout deal-terms database before investing outreach effort.
• Advisor, not founder. A named inventor stays part-time, and advisors should hold 10 percent or less. Check for a named non-academic cofounder and a full-time commitment signal.
• Timeline over-precision. No single standard formation duration exists and the stages overlap. Treat the 6 to 18 month window as a range, not a date.

The equity trap deserves its own number because it reorders rankings. Average university equity is about 5.9 percent in the US, 7.3 percent in the EU, and 19.8 percent in the UK, with a spread from 0 to roughly 70 percent in the open records. That means the same quality researcher is a materially worse deal on cap-table mechanics in a high-equity jurisdiction. Rank by institution policy, not just by the science, and you will avoid courting teams that cannot be funded.

For context on what a workable cap table looks like, benchmark guidance recommends 10 to 25 percent university equity for foundational-IP spinouts, and for software spinouts founders keeping 90 to 95 percent against a university 5 to 10 percent. A deal far outside those ranges is a flag to resolve before, not after, you build the relationship.

What to verify before you call the list done

Run this check before any name reaches an investor. It is the difference between a defensible watchlist and a list of professors.

Subject: Your work on [paper or repo topic]

I read your [paper and repo] and noticed you were both corresponding author and the main committer, which usually means you did the engineering rather than advised it. I invest at the point researchers are deciding whether to commercialise, before a round forms. No ask here, just an offer to compare notes on the spinout path from your institution if it is useful. Open to 20 minutes in the next few weeks?

Keeping the watchlist current

A watchlist decays because the signals keep moving: new camera-ready repos land, provisional filings surface, and I-Corps cohorts publish on a rolling basis. Re-run steps two through four on a cadence that matches your field's publishing rhythm, quarterly for fast ML venues and semi-annually for hardware fields tracked through patent classes. Re-date every outreach window on each pass, because a name that was 18 months out last quarter may have moved inside the SBIR gap.

Watch for three transitions that should bump a name up the ranking immediately: a provisional patent appearing under an existing author, an I-Corps seat added to someone you already flagged, and a university affiliation dropping off a profile, which often precedes a departure. The data lag that gives you an edge on awards works against you on departures, so the profile change is frequently the earliest readable sign that the window is closing. When you see it, move that name to first contact ahead of your scheduled cadence.

Finally, keep the disqualifiers live. Institution equity policy, cofounder status, and full-time intent all change, and a name that was unfundable on cap-table mechanics last year may become workable if the institution updates its spinout terms. Re-check the policy on the top of your list before each outreach cycle, and you will keep spending investor time only on the researchers who can actually be backed before they leave the lab.