The Direct-Secondary Holder Playbook: Transfer Verdict to Cleared Queue

This playbook is for early-stage investors, platform and talent partners at funds, and angels who want to buy common stock in one named late-stage private company directly from the people who hold it. Starting from the company name, it gets you to a documented transferability verdict within one working day. If the company is open, it then gets you to a queue of 25 to 50 former employees, each with a vesting band, a timing flag, and a first-contact script that compliance has cleared. Most secondary guides are written for sellers. This one is for the buyer who has to find and qualify them.

What has to be true before you source a single seller?

First, the shares have to be legally transferable to you, and you need to know that before you contact anyone. Several large private companies now publish notices saying any transfer without board approval is void, and when a trade is void, the buyer takes the loss.

OpenAI and Anthropic have both published these policies, and both name the same list of targets:

• Direct sales of shares
• SPVs (special purpose vehicles, entities that pool many buyers behind one holder of record)
• Tokenized interests
• Forward contracts (agreements to deliver shares later, usually after a liquidity event)

OpenAI's notice is blunt: "Any attempted transfer - which includes any pledge, encumbrance or other similar disposition - that does not follow this requirement is void." Anthropic went further and published a blocklist of unauthorized platforms that included well-known secondary marketplaces. So a trade can be blocked even when it runs through a marketplace.

The legal distinction matters. A voidable transfer can be unwound. A void transfer never happened. Gabriel Shapiro's reading of these policies is that "Sellers can retain both the shares and the money received, while buyers can only seek recourse upstream." All the risk sits with you. A single missed policy page can wipe out a whole sourcing pipeline, which is why every step about people comes after the steps about the company.

A void transfer can leave the seller holding both the shares and your money.

Transferability is not one question. It is a stack of restrictions, and a "no" at any layer ends or reshapes the deal.

ROFR (right of first refusal) is the company's right, and often its investors' right after it, to buy the shares on the exact terms you negotiated. It does not stop a deal from existing. It decides whether you or the company ends up with the shares.

The procedure, start to finish

The method has eight stages. The first three settle transferability, the next three build and rank the queue, and the last two clear and send outreach. Expect about one working day to reach the verdict, roughly two more days to a ranked queue, 2 to 5 days of compliance review, and then a ROFR window of typically 30 to 90 days once the seller files a transfer notice.

Some practitioners check tenders after sourcing. I put the check before sourcing because a tender changes how urgent each holder is. Someone who can sell into a tender next quarter is a different prospect from someone who has been shut out of every company program.

The transferability verdict: open with ROFR, tenders-only, or stop

The verdict has three possible values and comes from a half-day policy sweep plus a two-hour tender check. The governing documents run in parallel and add precision. They do not hold up the verdict unless they turn up a restriction.

The policy sweep (half a day)

Do not limit the search to the investor relations page. These policies show up on support pages and in press coverage, so a clean website search can be a false negative. Run at least these searches:

• The company name plus "transfer restrictions" plus "void"
• The company name plus "SPV" and plus "secondary"
• The company's help or support center, searched for "shares", "transfer", and "tokenized"
• Press coverage of the company's tenders, which often mentions transfer policy in passing

What it proves: a published notice is the company stating its position, and it is decisive. When it lies: finding nothing proves only that you found nothing. Screenshot every search result page with the date visible, including the empty ones, so the file shows what was checked and when.

The governing documents (parallel, 1 to 3 days if anything is public)

Counsel checks any filed charter or ROFR and co-sale agreement. Where these appear as exhibits to public filings, they show the real timing terms. Most late-stage private companies will have nothing public, and "not public" is an acceptable logged value. What you cannot do is assume the default terms.

The defaults themselves are disputed, as the table shows.

One secondary account says SpaceX's company window is 30 days and sometimes runs to 60. That claim is unverified, so treat it as a reason to plan for the long end, not as a fact about any particular company.

The three verdicts

Price ROFR as a real cost, not a rare event

In Hiive's platform statistics, ROFR was exercised on 18% of direct transfers in one year, up from 12% the year before, against a 72% company approval rate on direct transfers. Add the 10% that were canceled or otherwise not approved, and 28% of direct transfers that year did not deliver shares. You will see a general claim that 80 to 90% of secondary transactions end with the company waiving ROFR, but it comes with no methodology. Build your model on the documented figure.

One pattern deserves watching. ROFR exercise rose in the same period that Carta's tender count rose 62%. Companies running their own liquidity programs may have more reason to buy back blocked shares. That is a correlation, and the mechanism is my inference, not an established fact.

Company:
Date of sweep:
Swept by:
Searches run (exact strings):
  - [Company] transfer restrictions void
  - [Company] SPV
  - [Company] secondary
  - Support center: shares / transfer / tokenized
Notices found (URL + screenshot file name, or "none found"):
Governing documents: [ROFR cascade and notice periods] OR "not public"
Tender in last 6 months: yes / no
Ex-employees eligible in last tender: yes / no / unknown
VERDICT: void-policy (stop) / tenders-only / open with ROFR
Reviewed by counsel (name, date):

Do former employees get to sell in company tenders?

Usually on worse terms than current employees, and often not at all. In the median tender administered on Carta over a three-and-a-half-year span, no former employees were eligible. That exclusion is exactly why former employees are the core population for a direct buyer.

A tender offer is a company-organized sale, often a buyback, in which eligible holders can sell a capped share of their stock. Tenders are now common. Carta counted 396 in one year, up 62% on the year before, and about 20% of the employees who sold in them were former employees of the company running the tender. The median buyback caps employee sales at 25%, and former employees can sell nothing in the median buyback.

Eligibility rules can also be targeted. Rippling was reported to have barred former employees working at eight named competitors from its tender. Generic tender data will not tell you whether a given holder qualifies.

At the very top of the market, demand outruns supply. In one OpenAI tender, employees sold roughly $6.6 billion of stock and left $4 billion of buyer interest unfilled. If your target is in that tier, expect former employees to have choices and to be selective about buyers.

How tenders are announced to holders is not publicly established. In practice it happens through internal emails and equity portals. Record ex-employee eligibility as "unknown" until you have a press report or another direct source, and never let "unknown" quietly turn into "no".

Building the holder queue from public tenure records

Build the queue from employment history with dated end dates, not from "ex-" headlines. Aim for 25 to 50 rows, each with join date, leave date, and the title the person held at the target. Self-labeled profiles undercount the real pool and are polluted with people who never left.

Here is what self-labeled former employees look like in Refolk's index of professional profiles.

Three things follow from these figures.

• Seniority is concentrated. 429 of the 1,428 US ex-Stripe profiles (30.0%) carry Director, VP, CXO, or Owner titles. Senior former employees are a sizable slice, not a handful.
• Geography shrinks the pool fast. The UK count is 158, or 11.1% of the US count. If you are restricted to one jurisdiction, expect a far smaller queue.
• Pools are comparable across large late-stage companies. Ex-Databricks US headlines come to 1,263, or 0.88x the ex-Stripe count, so a 25 to 50 row target is realistic for companies of that size.

Each row needs these fields:

• Name and profile link
• Title held at the target (not the current title)
• Join date and leave date at the target
• Evidence source for tenure and title
• Current employer, with a flag if it is a competitor of the target
• Location, for jurisdiction checks

The "ex-" headline as a signal: it proves the person identifies with the company. It lies when the person never left, which is true of 20% to 24% of the sampled profiles, or when the headline marks an internship or a short contract. The fix is to search employment history directly for people with an end date at the target. Refolk takes that request in plain English, such as people who held a role at a named company for a set number of years and then left, and returns profiles matched on employment history rather than on headline keywords. That removes the step of collecting "ex-" profiles and then throwing out the fifth that are still employed.

Vesting bands and timing flags

Every row gets two labels: a vesting band (low, medium, or high, with a confidence rating) estimated from tenure, level, and stage at join, and a timing flag (now, wait, or skip) drawn from exercise-window and tender dates. Together they rank the queue.

The vesting band

Inputs, in order of weight:

1. Tenure at the target, measured against a standard multi-year vesting schedule
2. Title held at the target, which sets the likely grant size
3. Company stage when the person joined, since earlier joiners usually got larger percentage grants

The public benchmarks are thin. Carta reports a median first-engineer grant of 1.54%, with an interquartile range (the middle half of observations) of 0.61% to 3.5%. Those are startup grants. Level-by-level tables for late-stage companies exist only in secondary aggregators that cite their sources inconsistently. How far a tenure-based estimate lands from actual holdings is not publicly established. Report bands, give each a confidence rating, and never put a share count or percentage in a row. If you have private comparables, such as offer letters seen in other deals, use them to adjust confidence, not to produce a point estimate.

The timing flag

The timing flag rests on the post-termination exercise window. Incentive stock options (ISOs) keep their tax treatment under IRC Section 422 only if they are exercised within 90 days of leaving. Carta notes that three calendar months works out to 89 to 92 days depending on the time of year. Exercising ISOs can trigger alternative minimum tax on stock that cannot be sold. A holder in their first 90 days after leaving therefore needs cash at the same moment they face a deadline.

Extended windows remove that pressure. When a company extends the window, the options become non-qualified stock options (NSOs), but the holder keeps the right to exercise and loses the day-91 deadline. Pinterest was reported to have extended its window to seven years for employees with at least two years of tenure. Public lists of companies with extended windows exist, including the holman extended-exercise-windows repository on GitHub, Holloway, and Carta. The GitHub list has not been updated in years, so treat it as a lead to confirm, not a source.

Apply the labels this way:

• Now: left within the last 90 days with no known extended window, or left earlier, holds exercised stock, and is shut out of tenders.
• Wait: likely eligible for a confirmed upcoming tender, or still deciding whether to exercise. Recheck on a fixed date.
• Skip: still employed at the target, no end date on the profile, or a band too low to justify the transaction cost.

Compliance clearance and first contact

Every message is 1:1 to a holder you have pre-qualified, counsel signs off the script before the first send, and the buyer is documented as accredited. Allow 2 to 5 days for review, and do not let deal pressure cut it short.

Under the Section 4(a)(7) resale exemption:

• Each purchaser of the resold securities must be an accredited investor.
• The securities may not be offered through general solicitation.
• The seller must provide information such as updated financials.

The SEC also notes that a broker's Section 4(a)(4) exemption is not available when the broker knows, or has reasonable grounds to believe, that the seller's side of the transaction is not exempt. Intermediaries cannot look away from how the seller qualifies.

Some things this playbook cannot settle, because they are not publicly established in a form I can cite. These include exactly what triggers broker-dealer registration for a buyer who sources sellers, what confidential company information a seller may share, and how individual companies bar current employees from selling. Put them to counsel directly. Latham and Watkins' paper on legal issues for broker-dealers in private secondary trading is a useful starting point for that conversation.

Hi [first name],

I saw you were at [company] as [title at company] from [start year] to [end year].

I am an accredited investor and I am writing to you individually, not as part of a broad offering. I am interested in buying a small number of [company] common shares from former employees, if you hold vested and exercised stock and are open to selling.

Any purchase would go through [company]'s transfer process, including its right of first refusal, so the company would need to approve the transfer. Please do not send me any confidential company information.

If this is of interest, I am happy to set up a short call. If not, no reply is needed and I will not follow up more than once.

[Your name]
[Fund or entity name]

Once terms are agreed, the deal moves into the company's process. Most of the elapsed time sits here, and so does most of the risk of not delivering.

How this goes wrong: failure modes and false positives

Most failed direct-secondary pipelines fail in the same handful of ways. Each one below names what goes wrong, why it fools people, and the check that catches it.

• Missing a void notice. A clean search of the investor pages is a false negative, because these policies appear on support pages and in press coverage. Check: search the company name plus "transfer restrictions" plus "void", search the support center, and screenshot every result page with the date. If you miss this, the buyer bears the loss, because under a void policy the seller can keep the shares and the money.
• Keyword-matched holders who never left. In Refolk's index, 20% of the sampled ex-Stripe US profiles and 24% of the sampled ex-Databricks US profiles still list that company as their current employer. Check: require an end date at the target on the profile itself. A headline is not evidence.
• Founder titles inflating the band. In the Director-and-above ex-Stripe sample, 28% are now founders or CEOs. A current title says nothing about the level someone held at the target, and a founder title invites a high band nobody earned there. Check: score only the title held at the target, taken from the dated employment entry.
• A vesting band presented as precise. The only public benchmarks are startup grants, such as the 0.61% to 3.5% first-engineer range, and the error of tenure-based estimates is not publicly established. A share count in a row looks like diligence and is really a guess. Check: bands with confidence ratings only, and make sure nobody upstream turns a band into a price.
• Misreading the 90-day flag. A holder inside the window may hold unexercised options, not stock. Holders under an extended window, such as Pinterest's reported seven years, have no deadline at all. Check: confirm exercise status in the first conversation, and check the company against the public extended-window lists before you label anyone "now".
• Assuming tender eligibility. The median Carta buyback excludes former employees entirely, and Rippling excluded people who had joined competitors. "Wait for the tender" sounds patient and can turn into a lost deal. Check: confirm eligibility from a direct source before you use a tender as the reason to delay. Otherwise the flag stays "unknown".
• A script that counts as general solicitation. Templated mass outreach can break 4(a)(7) and put the whole trade at risk. Check: 1:1 messages to pre-qualified holders only, a compliance sign-off on file, and a send log showing who received what and when.
• Pricing ROFR as rare. The "80 to 90% waived" figure has no methodology. Hiive's documented non-delivery rate is 28%. Check: size the queue and your capital commitments on the assumption that more than one deal in four may not deliver.

Keeping the queue current

A holder queue goes stale quickly. Exercise windows close, tenders get announced, and companies publish new policies. Re-run the policy sweep before every term sheet, refresh timing flags on a fixed weekly cadence, and re-source when a top-quadrant row drops out.

Three dates drive the refresh:

• Each row's departure date plus 90 days. That is when a "now" row usually becomes a "wait" or a "skip".
• Any announced or reported tender. It can move rows from "now" to "wait", and it can signal that the company is about to tighten its stance on direct transfers.
• The date of your last policy sweep. A verdict is a snapshot, so date it and treat anything older than your current deal cycle as expired.

Departures keep happening, so the "left in the last 90 days" slice refills on its own. Re-running the same plain-English search in Refolk each week picks up new leavers without rebuilding the list. Use the checklist below before you call a queue ready to send.