Grading a Founding Team on Public Evidence

This guide is for early-stage investors, platform and talent partners, and angels who need to form a defensible view of a founding team before they spend partner time or ask for references. It isolates one repeated judgement call: scoring a founding team from public evidence alone, arriving at a go/hold/pass decision, and knowing exactly which claims to verify later. It is not a full diligence checklist and not a reference-check how-to. It is the scoring model for the moment before those.

The existing public material splits into two piles. One is process diligence that assumes you are already deep in a deal. The other is reference-check advice that assumes you are at term sheet. Neither gives you a repeatable way to grade two teams the same way at the earliest gate, using only what the open web will tell you. That gap is where conference-room gut feel takes over, and gut feel is exactly where pedigree bias lives. This framework replaces it with named, weighted dimensions.

Why score the team at all, and how much weight it deserves

The team is the single most cited factor in venture decisions, but it is a predictor of funding more than of outcome, so you score it for the meeting and verify it for the check. In the most-cited survey of venture investors, the management team was named an important factor by 95% of firms and the most important factor by 47%, ahead of business model at 83%, product at 74%, market at 68%, and industry at 31%.

That would argue for weighting team above everything else. A newer study of over 8,000 sourced deals complicates it: team predicts which companies raise initial funding, while market and product better explain larger financings and long-term success. In that dataset roughly 30% of sourced deals raise at least $1M. The lesson is not to demote the team. It is to cap its weight so that a strong founding narrative cannot drag a weak market through your gate. Practitioner framing rarely lets team fall below 30%, on the logic that strong founders course-correct when data changes. Keep it heavy, keep it bounded.

There is a cheaper reason to score the team early. Cofounder conflict is the load-bearing human failure: in Wasserman's work, 65% of high-potential startups fail from conflict among cofounders. That is a team-scoped statistic, not a general failure ranking. The general ranking looks different, and you should hold both in view.

The six dimensions and what weight each carries

Score every team on the same six dimensions, fixed in writing before you search any names. Freezing the rubric first is the control that stops a Stanford line or a famous logo from quietly reweighting your judgement mid-read.

Two dimensions deserve a specific caution before you weight them. Industry fit is the strongest predictor in the census-scale data yet is the one most rubrics never score explicitly, defaulting to "impressive resume" instead of "domain match." And cofounder alignment is the most predictive dimension that is the least publicly visible, which is why it must carry a low confidence flag rather than a high score.

What each public signal proves, and where it misleads

Every public signal is a proxy, and every proxy has a failure mode. Score the artifact, name it, and record what it would take to make it lie.

Domain and track record

Prior experience in the specific industry predicts much greater rates of entrepreneurial success, and the mean age at founding for the top 1-in-1,000 fastest-growing ventures is 45.0, not the mythical dropout. Serial-founder credit is real but non-linear. A Danish panel found serial entrepreneurs post 67% higher sales than novices, yet outcome-scoped figures show success peaks at the second venture and declines after: second-time founders at 34%, third-time at 28%, fourth-time and beyond at 22%. One US study even found that owners with a prior business had a 7% lower probability of exit. Read this as: weight domain fit over raw repetition, and stop giving linear credit for the count of past startups.

Technical build evidence

Build evidence is the one dimension the open web renders honestly, if you inspect substance rather than volume. High commit counts are theater when they are forks and trivial commits, so read authorship and the substance of what shipped. Patents are a genuine external signal: patent-owning startups are 6.4 times more likely to attract investment. Scarcity explains why this signal inflates. In Refolk's index, only 2.15% of US founders list machine learning as a skill, so credential and commit theater pays off precisely because real technical founders are rare.

Finding the small population that actually has the build history, rather than the population that claims it, is the friction. Describing the profile you want in plain English and pulling it straight from the open graph removes a day of manual disambiguation.

The verification pass: registries that prove existence, and only that

Verification means cross-checking every claim against a public record and marking it verified, unverifiable, or contradicted. These sources prove the existence or absence of a filed record. None of them capture private disputes, unfiled conflicts, foreign jurisdictions, or how someone actually performed inside a prior role.

Resume inflation is why this pass is not optional: roughly 33% of applicants admit lying about their education, and only 34% of employers verify the credentials on a resume. A LinkedIn title is a claim, not a fact. Route every unverifiable or contradicted claim to the verify-later list rather than scoring it as if it were confirmed.

Scoring a team from public evidence, step by step

Run the same procedure on every team so two deals are graded identically. The sequence below is one to two hours of analyst work per team plus review, and it ends in a one-page memo. Sources disagree on whether to verify before scoring capability or the reverse; I verify first because a contradicted claim changes what the capability score is even measuring.

How this goes wrong: the false-signal traps

Most bad team calls come from a short, well-documented list of traps. Each one is a false positive, so treat this section as the checklist you run against your own scores before the memo.

Pedigree substituted for capability. A Stanford or ex-FAANG line scores the team "strong" with no shipped artifact behind it. This is your default error: 70% of US VCs hold elite-institution degrees and demonstrably favor similar founders, and fund-level data shows one firm directing 67% of its seed investments to Ivy-League-Plus alumni and another 63%. Strip the school and employer credit, re-score on verifiable output only, and the bias becomes an edge, because it is a market-wide mispricing.

Media-list halo. "Forbes 30 Under 30" read as validation. Treat lists as unverified. Multiple honorees have faced fraud charges, including an AI startup founder, a Harvard spinout, charged with defrauding investors of $10M.

Unverifiable claim scored as verified. A title or date accepted because LinkedIn displays it. Mark it contradicted or unverifiable and route it to the verify-later list; roughly a third of applicants inflate their education.

Cofounder alignment inferred from absence of conflict. No public friction read as "aligned," when conflict is private by nature. Score alignment as low-confidence and require a downstream reference. This is the single largest failure cause at 65%.

GitHub or commit theater. High commit volume mistaken for capability. Inspect substance and authorship, not counts; forks and trivial commits inflate activity.

Serial-founder overcredit. "Third-time founder" scored top marks. Outcomes decline after the second venture from 34% to 28% to 22%, so weight domain fit over repetition.

Registry silence read as clean. No PACER or EDGAR hit taken as integrity, when absence of a federal record covers neither state courts, settled or unfiled disputes, nor foreign jurisdictions.

Pedigree bias is not just your error; it is a market-wide mispricing you can arbitrage with a rubric that scores verified output.

Turning the score into a call you can defend

The output is a weighted score converted into go, hold, or pass, paired with a list of exactly which claims references and formal checks must confirm downstream. Reference calls and formal background checks are not part of this gate. Formal diligence typically begins after investors express serious interest and runs two to six weeks, and reference calls are usually a term-sheet-stage activity. Your job here is to decide whether the deal earns that time, and to hand the downstream checkers a precise list rather than a vague "kick the tyres."

A defensible memo does three things: it states the weighted score and the artifact behind each dimension, it names the load-bearing unverified claims, and it says what the go/hold/pass call would flip on. Use this skeleton verbatim.

TEAM: <company> / founders <names>
CALL: Go / Hold / Pass

SCORES (weight x score):
- Domain / industry fit: __  (artifact: ____________)
- Founding-team track record: __  (artifact: ____________)
- Technical build evidence: __  (artifact: ____________)
- Cofounder alignment: __  [LOW CONFIDENCE]  (artifact: ____________)
- Integrity / verification: __  (registries checked: ____________)
- Market pull: __  (artifact: ____________)

WEIGHTED SCORE: __ / 100
DEBIASED SCORE (pedigree + media credit stripped): __ / 100

LOAD-BEARING UNVERIFIED CLAIMS:
1. ____________  (route to: reference / formal check)
2. ____________  (route to: reference / formal check)

THIS CALL FLIPS IF: ____________

Two markets frame how much supply you are choosing from, which matters when a partner asks whether this team is the best available rather than merely acceptable.

Against that supply, the framework is a filter, not a verdict. It advances or stops a deal and tells the next stage what to prove.

Keeping the framework honest over time

Re-run the debiasing control on your own past calls, not just the live deal, because that is how you learn whether your rubric is measuring capability or prestige. Before you sign off any grade, work the checklist below.

To keep the model current, track two things rather than any single number. First, whether your go calls survive the downstream reference and formal checks; a pattern of contradictions there means your verification pass is too shallow. Second, whether your debiased scores diverge from your raw scores in a consistent direction; if stripping pedigree keeps dropping the same teams, your rubric is still leaking prestige credit despite the frozen weights. The evidence base itself shifts, so re-check the mechanism, not the figure: when a new dataset revises how much team predicts outcome, adjust the cap on team weight, not the whole framework. The dimensions are stable. The weights are the thing you keep honest.

Refolk removes the two most manual parts of this work: disambiguating founders to canonical public profiles and finding the narrow, verifiable populations, such as founders with granted patents or prior senior roles at a named lab, that your build-evidence and domain scores depend on. The judgement stays yours; the sourcing and disambiguation stop eating the hour before the partner meeting.