--- title: Holder retention curves (Labs) summary: For each collection we follow, how long items stay at the address they were minted to, as a Kaplan–Meier curve with the share at 30, 90 and 365 days. published: 2026-10-08 updated: 2026-10-08 --- ## In one sentence For each collection we follow, the share of items still at the address they were minted to, plotted against days since the mint as a Kaplan–Meier curve; the first move counts whether or not it was a sale. ## What it is A Labs experiment. For each collection we follow, the share of its items still at the address they were minted to, as time passes since the mint, worked out from our own Bitcoin node. **"Left the address it was minted to" is not the same as "sold".** Moving an item to your own other wallet looks the same on-chain as selling it. The curve measures the first move, whatever it was. ## Where an item was minted to An inscription id is `i`. We read the reveal transaction from our node and find envelope `n`: envelopes are numbered in order through the transaction's inputs, the way ord numbers them. The inscription's sat is the first sat of the input carrying the envelope, unless the envelope names a pointer inside the outputs. Whichever output holds that sat is the **genesis output**, and its address is the address the item was minted to. If the envelopes can't be counted without doubt, or the sat goes to the fee, we don't guess: the item is left out and counted as unresolved. ## When it first moved Bitcoin Core doesn't index which transaction spent an output, so we read blocks. Every input of every transaction is checked against the genesis outputs we are watching. The block that spends one is when that item first moved. A move in the same block as the mint is found too. The walk starts at the newest block and reads backwards to the earliest mint of the collections we follow, a few blocks a minute so it never slows the node down, and keeps up with each new block. It only runs once our ord index has caught up. ## The curve For each item: days from the mint block to the block that first moved it. An item that hasn't moved has been watched until the newest block we have read; it is **censored** there, because we can't say anything about blocks we haven't read. The Kaplan–Meier estimate handles items watched for different lengths of time. At each time some items moved, with *d* moving and *n* still at their minted-to address just before: S(t) = product over move times up to t of (1 − d ÷ n) **Worked example.** Five items: one moved on day 10, one watched 20 days without moving, one moved on day 30, one on day 40, one watched 50 days without moving. Day 10: 1 − 1/5 = 80%. Day 30: three still watched, so 80% × (1 − 1/3) = 53%. Day 40: two still watched, so 53% × (1 − 1/2) = 27%. ## The 30, 90 and 365-day figures Each comes with how many items are old enough to count: those that moved before that day, plus those watched at least that long. A figure with fewer than 5 such items, or past the longest time we have watched anything, says "not yet" instead of a number. ## Coverage Until the backward walk reaches a collection's earliest mint, an item minted before the lowest block we have read could have moved in a block we haven't read. Those items are left out, and the collection is labelled **"partial since block N"**: the figures cover items minted since that block. ## Launchpads If one address received more than half of a collection's genesis outputs, the collection went out through a launchpad or the creator's own wallet, and "left the address it was minted to" would measure the hand-out, not the holders. Such a collection is marked **"distributed through a launchpad; not measured yet"** and gets no curve. ## Limits - A move is not a sale (see above). - One person with several wallets moving items between them counts as a move. - Only the collections we follow, and only items whose reveal our node can read. - A reorganisation of recent blocks is handled: the affected blocks are read again. ## The question Would retention change whether you buy into a collection? Vote on the Labs page.