The Co-Founder Durability Read: Backable, Watch, or Walk

You are about to spend an hour with a founding pair, and the biggest predictor of whether their company survives is not on their deck. It is the relationship between the two of them, and existing diligence habits only teach you to read chemistry in the room. This guide scores that relationship from findable public history - prior overlap, role complementarity, equity and leadership signals, tenure patterns - so you walk into the call with a verdict and a short list of gaps, not a blank page. It is written for pre-seed and seed investors, platform and talent partners, and angels who take the first call.

The output is one of three verdicts: Backable, Watch, or Walk, plus the three to five questions you still have to close in the meeting. The point is not to replace the meeting. The point is to stop the meeting from being the first time you think about the pair as its own scored object.

Why the pair deserves its own score

The founding relationship is the load-bearing part of an early-stage bet, and it is the part most diligence never scores directly. Guides tell you to read the individual founder - market fit, background, talent pull - and then to trust your read of the room. That leaves the single largest failure mechanism unscored.

The headline number here needs a caveat before you ever use it. Noam Wasserman, in The Founder's Dilemmas, is cited for the claim that 65% of high-potential startups fail due to conflict among co-founders. That figure traces back to a 1989 survey of investor perception, not a controlled audit of why companies actually died. Treat it as directional, not causal. A more conservative and better-grounded figure comes from CB Insights' startup post-mortems, which attribute roughly 23% of failures to "not the right team," with co-founder conflict chief among them.

Either way, the relationship matters enough to score, and the stage where it matters most is exactly where you have the least corroborating evidence. Standard scorecard weighting puts team quality at 30%, the highest single factor, ahead of market opportunity at 25%. At pre-seed, the team is the single most important factor because there is no traction to check it against. Investors spend about 15% of deck-review time on the team slide, the highest of any stage. That is the paradox: you lean hardest on the team read precisely when you have the thinnest data. A structured public-history read is how you add data before the call.

The four dimensions to score

Score a pair on four dimensions you can read from public evidence: role complementarity, leadership clarity, the equity signal, and the tenure timeline. Each proves something specific, and each has a way of lying to you.

Complementarity is the most consistent positive signal in the literature. Startups with complementary co-founders show a roughly 30% higher success rate. What it proves: the pair covers more of the build/sell/fund/operate surface than either could alone, and they are less likely to fight over the same territory. What it looks like when it lies: mere personality difference dressed up as complementarity. "One is loud, one is quiet" is not complementarity. Require distinct functional ownership, not vibe.

Leadership clarity tells you who breaks a tie. The larger-equity co-founder is usually the CEO, and across founding teams there is typically a designated lead who receives an outsize share of equity. What it proves: the pair has already answered "who decides" before you asked. What it looks like when it lies: two identical "Co-Founder & CEO" titles, or a CTO who quietly controls the board while the CEO is the public face. Cross-reference title against equity share and who signs filings.

The equity signal is subtle, because the ratio matters less than the record of the guide. The durable signal is whether the split was negotiated, not what number it landed on. What it proves, when readable: that the pair priced each person's contribution early. What it looks like when it lies: a clean 50/50 that was actually a default nobody argued for - and you cannot tell the two apart from public records.

The tenure timeline tells you what happened last time. What it proves: whether this pair has co-built before and how it ended. What it looks like when it lies: a polished profile that hides a prior messy split. Search for the earlier venture's outcome, not just its existence.

Reading complementarity and the leadership signal

Complementarity means distinct functional ownership across build, sell, fund, and operate; leadership clarity means one person visibly holds the tie-breaking authority. Both are readable from titles, filings, and who speaks in public, with cross-checks to keep titles from fooling you.

The cleanest complementary pair is functional: one builds, one sells or raises. To score it, assign each founder a single primary function from public evidence. A pair where both map to "build" is a duplication risk regardless of how well they get along. A pair split cleanly across engineering and go-to-market is the structurally stronger shape, and it is what several of the strongest founding pairs in Refolk's index look like when you filter for them - one engineering, one sales or growth.

There is a supply fact that should shift how you weight the technical seat. In Refolk's index there are 18,594 US "Co-Founder & CEO" profiles against 4,914 "Co-Founder & CTO" profiles - a 3.8:1 ratio. The genuine technical co-founder is the scarcer seat.

The practical reading: a pair with a real technical co-founder, not a CEO plus a contractor wearing a CTO title, is structurally rarer than the deck volume suggests. Weight it up when it is genuine, and check hard when the "CTO" has no engineering track record.

For leadership, the equity ratio is your most readable proxy for who decides, because the larger holder is usually the CEO. But that proxy is weakening. The median two-founder split moved from 60-40 in 2019 to 51-49 in 2024 on Carta. As gaps shrink, ratio tells you less about who breaks ties, so you lean harder on title, filing signatures, and who represents the company on panels and to press.

Reading the equity and vesting signal

The equity signal you want is evidence that the split was negotiated, not the size of the split. This is the single most misread part of founding-pair diligence, because the intuitive red flag - the equal split - is now near-mainstream, and the real risk is invisible in public records.

Start with what the norms actually are, so you stop penalising the wrong thing.

A 50/50 split is neutral. Nearly half of two-founder teams now split equally. The durable signal is negotiation: Wasserman's study of nearly 6,000 founders found teams who split equally by default, without negotiating, were roughly three times more likely to be unhappy with the split than teams who negotiated an unequal division. The act of arguing it out is what predicted satisfaction, not the final number. Speed compounds the risk: 73% of founding teams split equity within the first month, which is fast enough that many splits are defaults rather than decisions.

You cannot see whether a split was negotiated from a cap table. So the equity signal on public evidence has a ceiling: you can note the ratio, but the negotiation question is always a call item.

A public 50/50 split is not a red flag. An unnegotiated split is, and you can only find that out by asking.

Vesting is the same problem, sharper. The standard is a four-year vesting period with a one-year cliff, starting at incorporation, documented in a stock-restriction agreement with a company repurchase right. Its function is directly anti-breakup: it stops a departing co-founder from walking away with dead equity. Double-trigger acceleration is the recommended standard because it protects founders without scaring off acquirers. All of this lives in private agreements and cap tables, not public filings. For almost any pair, the honest public reading is "not established publicly."

This is why a pure public read tops out at Watch on vesting alone. It is the highest-leverage item you cannot see, so it is a permanent line on your call list, never a public Backable.

The scoring procedure

Run the pair through seven steps, roughly 90 minutes end to end, and end with a numeric score per dimension and a one-line verdict. The procedure is deliberately mechanical so two analysts scoring the same pair land close together.

Use a fixed rubric so the total means the same thing every time. Score each of the four dimensions from 0 to 3, for a maximum of 12.

Complementarity   0 = roles duplicate | 1 = partial overlap | 2 = mostly distinct | 3 = clean build/sell/fund split
Leadership clarity 0 = both claim lead | 1 = ambiguous | 2 = lead named, thin proof | 3 = lead clear across title, equity, filings
Equity signal      0 = default split, no evidence of decision | 1 = ratio known, negotiation unknown | 2 = ratio + some negotiation signal | 3 = documented negotiated split
Tenure timeline    0 = prior split ended badly | 1 = no shared history, unproven | 2 = clean shared history, some overlap | 3 = complementary shared history, prior win
TOTAL (0-12): ___
Verdict: 9-12 Backable (pending vesting) | 5-8 Watch | 0-4 Walk

Note the ceiling on the equity dimension. Because negotiation and vesting are rarely public, most pairs cap at 1 or 2 there until the call. That is by design: a pair should not read as Backable on paper for the one thing you genuinely cannot verify from the outside.

When you need to build a set of pairs to run this read against rather than react to inbound decks, describing the shape you want in plain English is faster than filtering profiles by hand. Refolk returns the pair, not just one name, so you can score the relationship from the first pass. It is also where the supply figures in this guide come from: the 3.8:1 CEO-to-CTO ratio and the geographic concentration of co-founder profiles in the Bay Area, New York, and London.

How this read goes wrong

Most bad durability reads come from a small set of repeatable errors, and each has a specific check that catches it. This is the part of the standard worth re-reading before every call, because the failure modes are seductive - they all feel like rigor.

Treating 65% as a hard failure number. The false positive is citing it as causal proof that conflict kills two-thirds of startups. The check: it is an investor-perception survey from 1989. Pair it with CB Insights' more conservative 23% "not the right team" figure, and never present it as an audit.

Scoring shared history as automatically positive. The false positive is "they worked together for a decade, so they must be durable." The check: the exit-outcome study of about 350 US tech IPOs and $1B+ in exits found founders who had worked together before produced 21% lower exit valuations, and same-school pairs 7% lower. Convenience pairing skips complementarity. Score prior collaboration as a plus only when the pair is also functionally complementary.

Reading 50/50 as a red flag by default. The false positive is penalising any equal split. The check: equal splits are now near-mainstream at 45.9%. The real risk is an unnegotiated default, which you cannot see publicly, so ask in the call.

Assuming vesting is in place because it is standard. The false positive is scoring a pair as protected when no repurchase right actually exists. The check: vesting lives in private agreements. Mark it "not established publicly" and confirm the four-year, one-year-cliff schedule and double-trigger acceleration in the meeting.

Inferring a lead from title alone. The false positive is two "Co-Founder & CEO" titles read as shared strength, or a CTO who actually controls the board read as second fiddle. The check: cross-reference equity share and who signs the filings.

Confusing complementarity with personality difference. The false positive is "one is loud, one is quiet, they balance each other." The check: require distinct functional ownership - build, sell, fund - not temperament.

Over-indexing on a clean public profile. The false positive is a polished LinkedIn that hides a prior messy split. The check: search for the earlier venture's outcome and any litigation, not just its name.

What to carry into the call

Close the read by turning the score into a verdict and a named question list, so the meeting tests exactly the gaps public evidence left open. A durability read that does not hand you sharper questions has failed at its only job.

The three questions that survive almost every read, because public records rarely answer them, are: what are the vesting terms and is acceleration double-trigger; who has final decision authority when the two of you disagree; and if either of you co-founded before, why did it end. Lead with those. They convert a paper score into a real durability read faster than anything the deck contains.

Keep the read current by re-checking the mechanism, not the number. The equity norms in this guide will drift - the median split has already moved from 60-40 to 51-49 in five years - so before you lean on a ratio to infer leadership, confirm the current spread from the latest Carta founder ownership data rather than trusting the figures here. The supply picture will move too: re-run the CEO-to-CTO and geographic counts against Refolk's index when you want a fresh sense of how scarce a genuine technical co-founder is in the pool you are sourcing from. The verdict framework holds; the calibration inputs do not, and a standard that pretends otherwise is worse than none.