Methodology

Holder overlap network (Labs)

0.9.53

Open the dashboard

Which of the collections we follow share holders, how many, and how much more often than chance would give, with custodians left out and counts only.

In one sentence

For every pair of the collections we follow, the holders they share are counted and compared with what chance would give (the lift), with custodians left out and pairs with fewer than 10 shared holders not shown.

What it is

A Labs experiment. For every pair of the collections we follow, we count the holders they share and compare that with what chance would give. The result is drawn as a network: a line between two collections means their holders overlap more than you would expect if holding one said nothing about holding the other.

Which collections

The ones whose owners the dashboard already re-reads every day: collections someone here holds or watches, and the 50 with the most sales volume our node saw in 30 days. A collection is included only once our own ord index has answered for at least 90% of its members, the same rule the holder counts on a collection page follow. The card says how many are in the network and how many are still waiting.

Owners come from our own ord index and move to the buyer the moment our node sees a sale. Nothing is read while ord is still catching up: an owner read during indexing is the owner as of the block ord has reached, which can be months ago.

The numbers

For collections A and B, among N distinct holders across every included collection:

A lift of 4.2 reads: holders of A are 4.2× as likely as chance to hold B. Lift is symmetric, so the same is true the other way round. A lift near 1 means no link beyond chance; below 1, the two collections' holders tend to be different people.

Worked example. 1,000 holders in all; A has 100, B has 50, and 20 hold both. Expected: 100 × 50 ÷ 1,000 = 5. Lift: 20 ÷ 5 = 4. Overlap: 20 ÷ 50 = 40% of B.

What is left out

• Custodians. Before counting, an address that holds items in more than half of the included collections is dropped from every one (the share is a setting). Exchanges, escrow and marketplace wallets hold a little of everything and would otherwise tie every pair together. The filter is applied only with 10 or more collections in the network: with three, "more than half" is two, and ordinary collectors would be dropped. The card says how many addresses were left out.
• Small pairs. Pairs with fewer than 10 shared holders are not shown (also a setting). With a handful of shared holders, lift is mostly noise.
• The drawing puts the collection you choose (search for it, or pick one on the circle) at the top of a circle, with every collection sharing at least the minimum number of holders with it round the rest, most shared holders first. A stronger line (thicker, more opaque) means more shared holders, in four steps whose counts the legend gives; the list under it gives the exact counts, and lift is listed below that. Collections on the circle that also share holders with each other are joined by fainter grey curves, on the same strength steps; pointing at a collection lights up its own curves, and a checkbox hides them all.

Limits

• One person with several wallets counts as several holders. We see addresses, not people.
• It refreshes once a day, after the daily owner reads, so it is yesterday's picture.
• Only the collections we follow. A strong link to a collection outside that set doesn't show.
• A shared holder could hold one item of each; the overlap says nothing about how many items they hold.

Privacy

Counts only. The network is worked out from addresses, but no address is stored with it or sent to the page, and there are no per-wallet suggestions ("you hold X, so…").

The question

Did this show you a collection you'd look at? Vote on the Labs page.

Also as plain Markdown. See every methodology page.

Last updated 2026-10-08