Sourcing the Off-List References to Call on a Founder

Before you wire, you want signal the founder could not stage. This is the sourcing procedure for that: how an early-stage investor turns a founder's public history into a ranked, contactable list of independent references - former co-founders, early employees who left, prior-company colleagues, co-investors, and customers - that the founder never handed over. The published guides cover the timing of reference calls and the red-flag question scripts well. They treat "find two or three independent references" as one throwaway line. This is that line, expanded into a method you can run start to finish.

Why off-list sourcing is the part that decides the outcome

The off-list ratio, not the raw call count, is the signal of a serious reference process. Named references are pre-briefed by the founder, so they move conviction only at the margins. The people the founder could not anticipate are where anything real surfaces.

The volume numbers make the point. Baseline practice is modest: CRV recommends five to seven conversations, and a common founder-side norm is that the VC calls three to five of the references offered plus one or two off-list found independently. Top-tier practice runs far higher and skews independent: firms including Sequoia, Andreessen Horowitz, and Benchmark are described as running 10 to 20 or more calls per deal, targeting people the founder names and people the founder does not, with most not founder-named. One fund, 81 Collection, states it does 10 back-references before investing. The recurring theme across sources is that most of the calls are not to people you were given.

So the question is not "how many calls" but "how many of them could the founder not stage." That is entirely a sourcing problem. If your list is eight names and six came from the founder's orbit, running 15 calls buys you very little. The work below is about making the independent fraction large and correctly ranked.

Named references are pre-briefed. Conviction only moves on the people the founder could not stage.

Reference-call volume and off-list posture by source

Different firm tiers run different volumes and different off-list postures. The table below is the map: find where your firm sits, then size your sourced list to match.

The pattern is consistent: as firms get more serious, the total climbs and the share the founder controls falls. Top-tier volume runs roughly two to four times the CRV baseline. Whatever your tier, work backwards from calls booked. To book 5 to 10 calls you generally need a sourced shortlist of 8 to 15 names, because scheduling and non-response thin the list before it reaches your calendar.

The four public sources and what each one proves

Four primary public sources reliably surface people the founder did not name. Each proves a specific kind of association, and each has a documented limit that tells you when it is lying to you.

• LinkedIn work-history overlap. Matches on company name, position, and the date range of tenure, sorted by strength of connection, meaning how long you overlapped. Proves co-tenure, not relationship quality.
• SEC Form D on EDGAR. Publicly filed within 15 calendar days of the first sale, disclosing company, location, amount raised, round type, and the names and titles of signing executives, plus investors when they are named. Proves prior co-investors and officers.
• GitHub contributors and commit co-authors. The contributor graph shows who committed to a repository, and co-authored commits are attributed via "Co-authored-by" trailers visible in the contribution graph. Proves who built alongside a technical founder.
• Prior investors' portfolio pages. Surface co-investors and peer founders from earlier rounds.

Two of these have a timing advantage worth naming. Form D must be filed within 15 days of first sale and is instantly public on EDGAR, so co-investor names surface before any press announcement. That makes prior-round co-investors contactable earlier than press-based tools reveal them.

Read the limit column as a warning about coverage. If a founder came from a company over 2,000 employees, the standard overlap method under-covers them by design, and you fall back to manual date-matching. That matters because large-company histories are exactly where a problematic prior role can hide in a crowd.

The end-to-end procedure

Run these seven steps in order. The whole sequence fits inside a normal diligence window: seed diligence runs two to four weeks, Series A four to eight weeks, with roughly 30 days of confirmatory diligence after the term sheet. Because calls run in parallel, referencing stays off the critical path to signing.

On step one, sources genuinely disagree on order, and I will not pretend otherwise. Some firms disclose backchannel upfront, putting it on a "how we invest" page shared before the first meeting; others begin quiet backchannel during the first or second meeting, before the founder knows. Disclosure is not only a compliance move - firms that disclose report it builds trust and balances the asymmetric power dynamic, which can improve candor rather than reduce it. Pick a posture deliberately and apply it consistently.

How to rank the shortlist by signal

Rank by incentive and independence, not by how easy someone is to reach. The person with the most to tell you is usually the one with the least reason to protect the founder's story and enough recent proximity to know the truth.

The highest-yield node is the early employee who left in the last year, found yourself on LinkedIn. Recent departures retain current knowledge but no equity-driven incentive to protect the narrative. This is not theory. In a documented case, an investor went off-list and found someone who had worked with the founder at a previous startup; that source said the founder had been fired for inflating user metrics in investor updates by 15 to 20 percent. Eighteen months later the SEC filed civil fraud charges, having found the founder booking test transactions as real customer activity. That signal came from exactly this kind of person, and no named reference would have produced it.

Use two axes to sort: how independent the person is from the founder's current incentives, and how directly they observed the founder's work. The people you want most sit in the top-right corner.

Alongside the recent-departure employee, CRV specifically weights the failed-venture reference: a co-founder of a company that struggled or did not reach its expected outcome. How someone behaves when things go wrong is harder to stage than how they look when things go well. Prior co-investors round out the set - they see governance behaviour and how the founder handled bad news across a full round.

This is the point where sourcing off-list references stops being a lookup and becomes a search over incentives and dates. Naming the exact people - by prior company, tenure window, and current employer - is the friction. Refolk is built for that: you describe the people you want in plain English and get them back across LinkedIn, GitHub, and the open web, which is precisely the enumerate-and-rank work in steps three and four.

How this goes wrong

The failure modes here are not edge cases; they are the default way the method underperforms. Each has a tell and a check.

• Overlap without acquaintance. Co-tenure at a large company proves nothing about whether two people ever spoke. Overlap tooling itself skips 2,000-plus-employee firms and applies a minimum overlap threshold for a reason. Check: confirm same team or office and a meaningful overlap length before you call.
• The coached "off-list" reference. A name you believe you found independently may still sit in the founder's orbit. The tell is uniformly glowing, specifics-free praise. Check: ask course-correction and forced-ranking questions; a three-second pause before answering is data.
• Survivorship bias. Calling only people still at the company or still friendly filters out the signal you came for. Check: force at least one departed-before-liquidity name onto the list, per the rule above.
• Treating a thin network as a red flag. Absence of a strong reference network is a false positive, especially for underrepresented founders building outside established networks. Check: weight a missing network separately from any negative signal you actually collect.
• Form D mis-attribution. The "executives" named on a filing may be lawyers or nominal officers, not co-founders. Check: cross-reference every title against LinkedIn tenure before treating a Form D name as a colleague.
• GitHub identity gaps. A real collaborator can be absent from the contributor graph if their commits used an unlinked email or landed on a non-default branch. Check: pull the commit's .patch to recover the author identity, and remember the graph lists only the top 100 contributors.
• Skipping lawful-basis documentation. Relying informally on legitimate interest fails an audit without a written assessment. Check: keep a one-page Legitimate Interest Assessment on file for each process.

A closing note on signal quality: the point of the call script is to defeat the coached reference. Forced-ranking questions ("where does this founder sit among the people you have worked with") and course-correction questions ("what did they get wrong, and how did they handle it") are hard to answer with rehearsed praise. Listen for hesitation as much as for content.

A copy-paste rubric and the outreach ask

Score each candidate the same way so the ranking is defensible when a partner reads the memo. The rubric below is deliberately simple; the discipline is in applying it to every name, not in its sophistication.

Independence (0 = current teammate, 2 = departed before liquidity or arm's-length co-investor): __
Direct observation (0 = adjacent team, 2 = worked directly with founder): __
Recency (0 = overlap ended 3+ years ago, 2 = left within the last year): __
Adversity exposure (0 = only saw good times, 2 = saw a shut-down, layoff, or missed target): __
Total (max 8): __
Reason to call, one line: __
Source and overlap window: __

Keep the outreach itself short, honest about who you are, and low-pressure. You are asking for a confidential conversation, not a testimonial.

Subject: Quick confidential question about [prior company]

Hi [name] - I'm [your name], an investor at [firm]. I'm doing diligence and I understand you worked with [founder] at [prior company] around [year range]. I'd value 20 minutes of candid, off-the-record perspective on what they were like to work with. Nothing you say gets attributed. Would a short call this week work?

Before you call the sourcing done, run the checklist.

Keeping the method current

Re-check three moving parts before each deal rather than trusting a saved playbook, because the coverage of your sources shifts. First, overlap tooling limits change: confirm the current result cap and employee-size threshold, since a founder from a large prior employer needs manual work whenever the automated method under-covers. Second, Form D timing is a mechanism, not a fixed date - filings are due within 15 days of first sale, so for a very recent prior round the co-investor names may not be public yet; re-query EDGAR closer to your decision. Third, treat the lawful-basis question as unsettled. There is no published investor-specific standard, so the safe move is a written assessment for each process and a consistent notice posture you can defend.

The exclusive comparison this guide would carry - how many former early employees from a given prior company have since changed employers, split by market - was not obtainable this cycle from Refolk's index. When you run your own sourcing, that departed-employee count in a market is the number that tells you how deep your off-list pool really is. Ask for it directly: name the prior company, the role, and the tenure window, and rank what comes back with the rubric above. The founder gave you their list. This is how you build the one they did not.