One firm, one record

“Index Ventures” arrives from one source. “Index Ventures Growth III” from another. “index.vc” from a third, and “Index Venture Associates VII Limited” from a companies register. A fourth source has “Index” with a London address and no website at all.

How many firms is that? The honest answer is that you cannot tell from the names, and a register that guesses will be wrong in both directions — merging distinct firms into one entry, or leaving one firm scattered across five.

Both failures are visible to the person using it. A merged record shows a cheque range spanning pre-seed to growth and a geography of everywhere, which matches every search and helps with none. A scattered one means the firm you were looking for was in the register the whole time, three rows away from where you looked.

Names are the weakest evidence available#

Name matching is the first thing anyone tries and the first thing that fails.

Legal names and trading names differ, and the register that holds one rarely holds the other.
Fund vehicles are named after their manager. Roman numerals and vintage years make a single manager look like a family of firms.
Suffixes — Capital, Ventures, Partners, VC — are close to noise. Two firms sharing one are not related; one firm may use it inconsistently across its own website.
Transliteration is not deterministic. A Greek or Japanese firm has several defensible Latin renderings, and different sources chose differently.

Normalising names helps at the margins and creates new problems at the centre: strip enough and “Alpha Capital” in Athens and “Alpha Capital” in Austin become the same string.

What identity is actually made of#

Something has to be stronger than the name, and in practice a small number of signals carry almost all of the weight.

A verified web domain#

A domain is the closest thing to an identifier a private firm has. It is chosen by the firm, it is public, and two firms cannot hold the same one. Where a domain is present and confirmed, it settles the question on its own.

Where it is absent — and for family offices, angel syndicates and older firms it often is — nothing else is nearly as decisive, and the bar for merging has to rise accordingly.

Jurisdiction and register identifiers#

A firm registered in one jurisdiction with a register number is pinned to that number. This is why register imports are worth the effort even though they carry almost no investment metadata: they contribute identity, and identity is the scarce resource.

Corroborating structure#

People, portfolio companies and addresses form a shape around a firm. Two records sharing partners and portfolio companies are far more likely to be one firm than two records sharing a normalised name.

Choosing not to decide#

The most useful state in a resolution pipeline is not “same” or “different” but “undetermined”. A pair that is plausibly one firm and cannot be confirmed should stay two records and be marked as a candidate, not resolved on a tie-break.

This sounds like an admission of failure and is the opposite. A register that resolves everything has necessarily resolved some things wrongly, and it cannot tell you which. One that carries its uncertainty explicitly can be checked, corrected and improved, and the person reading a record knows whether they are looking at a settled fact.

The same principle governs what a merge is allowed to do to the surviving record. Two records that disagree about a firm’s headquarters do not become one record with both headquarters, or one with whichever value happened to load last. The conflict is a fact about the sources, and hiding it produces a record that looks more confident than the evidence behind it.

Why this is on the blog rather than in the footnotes#

Because it explains the shape of what you see. Coverage is uneven, some records are thinner than others, and a firm you know exists may appear under a name you did not expect. That is not carelessness; it is what a register looks like when it declines to guess.

If you find a record that is wrong, split or duplicated, the correction routes are on the record itself — firms and individuals can claim, correct and request removal of what we hold about them.

Names are the weakest evidence available#

Name matching is the first thing anyone tries and the first thing that fails.

Legal names and trading names differ, and the register that holds one rarely holds the other.
Fund vehicles are named after their manager. Roman numerals and vintage years make a single manager look like a family of firms.
Suffixes — Capital, Ventures, Partners, VC — are close to noise. Two firms sharing one are not related; one firm may use it inconsistently across its own website.
Transliteration is not deterministic. A Greek or Japanese firm has several defensible Latin renderings, and different sources chose differently.

Normalising names helps at the margins and creates new problems at the centre: strip enough and “Alpha Capital” in Athens and “Alpha Capital” in Austin become the same string.

What identity is actually made of#

Something has to be stronger than the name, and in practice a small number of signals carry almost all of the weight.

A verified web domain#

Where it is absent — and for family offices, angel syndicates and older firms it often is — nothing else is nearly as decisive, and the bar for merging has to rise accordingly.

Jurisdiction and register identifiers#

Corroborating structure#

Choosing not to decide#

Why this is on the blog rather than in the footnotes#

If you find a record that is wrong, split or duplicated, the correction routes are on the record itself — firms and individuals can claim, correct and request removal of what we hold about them.