Who believed in you first?

Section 1The pitch, in their words

Cosign describes itself as a curated professional network for the people and companies that drive the startup ecosystem, made by Andreessen Horowitz and introduced by Erik Torenberg and David Booth on 25 September 2026. MTS described the same day as "Wikipedia meets LinkedIn meets AngelList." The homepage puts it this way: Believe in someone, first.

Build and browse profiles that showcase what makes someone exceptional, cosigned by the people who believe in them.

cosign.co, homepage, retrieved 27 September 2026 cosign.co

The public profile layer presents attributed professional endorsements from people who have worked together. The launch copy also describes an early-discovery angle. It does not establish how profiles are ranked, whether an endorsement changes visibility, or how recommendations work.

The launch materials also describe a separate intent network where members privately signal who they might fund or work with. That is a different kind of information from Form D's public issuer notice. Neither record can stand in for the other. MTS announcement Dealroom on the launch

Form D does not document endorsements or identify investors. It names certain people associated with the issuer and separately reports an investor total. This investigation follows that issuer-side record without treating it as evidence of who endorsed or invested in whom.

Section 2What Form D records

Rule 506 of Regulation D provides an exemption from registration when the issuer meets the rule's conditions. Under 506(b), the issuer may not use general solicitation or general advertising. It may sell to any number of accredited investors and up to 35 non-accredited purchasers in any 90-calendar-day period who, alone or with a purchaser representative, can evaluate the investment's merits and risks; the issuer may reasonably believe this condition is met. Under 506(c), general solicitation is permitted if every purchaser is accredited and the issuer takes reasonable steps to verify accredited status. See the SEC Rule 506 guidance and 17 CFR 230.506.

The SEC says a new Form D notice is due within 15 calendar days after first sale. If that deadline falls on a Saturday, Sunday, or holiday, it moves to the next business day. An amendment may be required for specific changes or annually while an offering continues. For this purpose, the first sale is when the first investor is irrevocably contractually committed. The notice is required, but the SEC staff says filing it is not a condition to the availability of the Rule 504 or Rule 506 exemption; Rule 507 describes potential consequences for noncompliance. See the SEC filing guide and the SEC's Form D FAQ and the SEC's Securities Act interpretations.

It asks for the issuer's related persons, including officers, directors and promoters, by name. It asks what industry it is in, what it is raising, how much it has sold, what the securities are, which exemption is being used, and the total number of investors it reported. The official form lists its fields. Here are five sample filings rendered from parsed data, so you can see what the same page looks like when the issuer is a biotech, an insurer, a property vehicle and a fund. The three rows with a warm tint are the ones this story turns on: what was offered, what was sold, and the reported investor total. The last of those is a count with no names attached, which is the focus of this chapter.

The cache preserves … sampled EDGAR index rows that explicitly mark the first sale as yet to occur. Other first-sale dates were not captured in the current cache, so filing cards label those dates unverified.

What the form gives you

It gives an issuer's related-person list, industry, offering amount, amount sold, securities, claimed exemption, investor total, and dates. Matching notices by CIK groups them by issuer, but does not establish that multiple notices belong to the same offering. Those fields support a directory, sector breakdown, and issuer-side name graph.

What the form does not give you

Form D reports an aggregate investor count in a separate item; it does not provide investor names. Its related-person list is not an endorsement list, and the filing does not say whether a listed person also invested.

Section 3Two thirds of issuers are funds

Loading the crawl.

Form D is filed for offerings under several exemptions, including Regulation D. In this sample of distinct issuer CIKs are pooled investment funds; the rest are non-fund issuers.

A non-fund issuer profile does not by itself identify a startup. Of the non-fund issuer profiles, fall into technology, health care, business services or manufacturing groups. The other include issuers in the broad “Other” group, property vehicles, banks, insurers, energy companies and restaurants. These labels alone do not establish whether an issuer is a startup.

One profile per non-fund CIK, searchable. CIK-attributed related-person name keys are unioned across sampled filings; other profile values come from its latest sampled row. If an accession appears under multiple CIKs, its amounts and investor total are joint-offering values that can repeat across rows. Its Item 3 names are shown separately in the details and are not assigned to each CIK. Secondary CIK names from the EDGAR index show an ellipsis and may be truncated; the index caps names at 55 characters. Use the details button to inspect the profile; issuer names link to EDGAR. A person's button filters to issuer profiles where that normalized name appears. A starred investor count of 0 may be an explicit zero or a blank the earlier parser stored as zero. Amounts are reported as sold in the latest sampled filing, not a valuation, and fund profiles report no amount in their latest sampled filing.

Section 4An issuer-related-person graph

This graph connects normalized names from CIK-attributed Form D related-person lists to issuer CIKs. Names from accessions listed under multiple CIKs are omitted: Item 3 may cover any issuer on the notice and the structured list does not allocate a person to a particular CIK. Those joint names remain visible in directory details without CIK attribution. The graph describes issuer-side filing roles; it does not show professional endorsements or investor identity.

CIK-attributed names reported on multiple profiles

The table lists normalized names reported on more than one non-fund CIK in accessions that can be attributed to the filing's primary issuer CIK. Many highly repeated entries have roles such as filing agent, general partner, administrator, or an officer of an issuer's administrator. The filing reports those issuer-side roles. Its separate investor total does not establish whether a listed person also invested.

Each repeated normalized person name contributes one link for each non-fund issuer CIK profile where it appears in a CIK-attributed filing. The table groups those links by the profile's latest sampled industry. It describes this sample's mix of issuer categories; it does not identify whether a person is a founder, investor, or startup operator.

Section 5The separate investor count

Form D has a box, near the bottom, that asks for the total number of investors who had already invested in the offering by the time the filing was made. The positive counts in the cache are known. An earlier parser stored blank fields as zero, so the cached zeros may be explicit zeros or prior blanks; their exact split is unknown because the cache was not refreshed. The chart below shows positive counts only. Among those rows, the distribution has a larger group at one and another at eleven or more.

This is a reported total for an offering. It does not identify investors or describe their allocations. It also does not show whether a person in the issuer-related-person list invested.

The amount-sold-per-investor statistic below is aggregate amount sold divided by the reported investor count. It is not an observed individual check size or a reconstruction of an issuer's capitalization.

Section 6A hypothetical network model

This section is a simulation, not a description of Cosign's actual product or user behavior. The checked launch copy supports attributed professional endorsements and an early-discovery angle; it does not document ranking, visibility thresholds, reputation weighting, or extra reach for early endorsers.

For a sensitivity exercise, the model assumes a set of people, some of whom can endorse; a visibility threshold; an endorsement weight based on the endorser's modeled reputation; a multiplier on endorsement weight for members of the day-one cohort; and traffic divided between discovery and social proof. Each day-one member starts with exactly the visibility threshold's number of endorsements, so they begin visible. Every modeled endorsement counts toward visibility; the seeded quality and noise draws decide whether it adds reputation. These are explicit model assumptions. Results describe what follows from these choices, not how Cosign ranks profiles or distributes visibility. The plotted runs share the same deterministic seed, quality draw, and day-one cohort; the dashed run changes discovery flow only.

Move each dial to compare outcomes inside the simulation. The ranges below report its parameter sweeps; they are not measurements of Cosign or claims about how its product works.

Section 7Receipts

Scope and limits

The SEC identifies Form D for exempt offerings under Rule 504 or 506 of Regulation D and Section 4(a)(5). Regulation A and Regulation Crowdfunding use other forms. The sample also cannot show offerings made entirely outside the United States or notices that are late or missing. A late or missing Form D can violate the notice-filing rule; the SEC staff says filing itself is not a condition to Rule 504 or 506 exemption availability, while Rule 507 describes possible consequences. See the SEC filing guide and its interpretations. Foreign issuers also appear in the data: non-US issuer CIKs are represented.

The graph groups filings by CIK and keeps one name entry per issuer, omitting shared-accession Item 3 names that cannot be assigned to a specific issuer CIK. Names are normalized for case, diacritics, and punctuation. For names matching the person-name heuristic, listed generational and honorific suffixes are removed only at the end. Entity-classified names keep every token, including Roman numerals. The method does not expand initials or aliases, and same-name matches are not verified identities. The current cache also omits the filed middle-name fields for related people; none were reconstructed. The corrected parser preserves middle names in future records, but this cache was not refreshed. Missing middle names can merge distinct people, and name keys can also split one person across variants; the net direction of name-count error is unknown.

The independent verifier recomputes data-derived counts from the cached filings and checks them against the generated page data. Model outputs are generated by the simulation. The code and raw cache are in the repository.