If the company running your Bitcoin wallet went offline tomorrow, could you still move your coins? For most wallets the honest answer is no, and everyone treats that as the price of convenience. Spark is interesting because it answers yes. That answer is the reason a growing number of Lightning wallets now run on it, and it is worth examining before a business trusts it with real money.
Spark is a Layer 2 built by Lightspark. Instead of Lightning channels it holds your Bitcoin in a structure the operators co-sign, and it hands you a pre-signed transaction that withdraws to the main chain without their involvement. That withdrawal, called the unilateral exit, is what separates self-custody from a custodial account. If it works when you need it, the operators can censor you or vanish and your Bitcoin is still yours.
Lightspark's material says the exit makes Spark self-custodial and your funds always recoverable. That claim is worth testing rather than repeating, so I went looking for what actually happens when you try to leave. The question that matters is not whether the exit exists. It is what has to be true for it to work.
What you are actually holding.
Your Spark balance is not a single coin. It is a set of leaves in a tree of pre-signed transactions, each leaf a small Bitcoin output sitting off-chain. Leaves split and merge as you pay, so one balance ends up scattered across many of them. In July 2026 the team at Blink published a real exit on Bitcoin's main chain, and their test wallet held 100,000 sats spread across 22 leaves. That number is the first sign that leaving is not one clean step.
The exit is paid for from outside Spark.
A unilateral exit means broadcasting each leaf's chain of transactions to the main chain and waiting out a timelock of roughly 1,400 to 2,000 blocks, about ten days to two weeks. The complication is the fee. Those transactions were signed in advance at zero fee, so you cover them with a separate on-chain input, money that cannot come from your Spark balance. To get your Bitcoin out, you need other Bitcoin already on-chain to pay the way out.
Blink's figures show why that matters. Four of their 22 leaves held 90 percent of the value. The other eighteen were small change from ordinary use, some as little as one sat, and each would have cost more in fees than it held. Exiting everything would have taken 253 transaction packages and about 78,000 sats in fees, roughly four-fifths of the wallet. Exiting only the leaves worth exiting recovered about 90 percent of the value for around 9,400 sats and a ten-day wait. The fragmentation Spark keeps out of sight during everyday use comes back at the exit, and it sets a floor on how small a balance is worth rescuing.
The exit depends on data you must already hold.
A pre-signed exit is useless without the recovery data for your current leaves, and this is where the promise narrows. Blink are precise about it: that data has to be fetched while the operators are online, and refreshing it takes the same cooperation as making a payment. Whatever your last refresh covered stays exitable. Whatever it did not, is not.
Put the two halves together. Fetching your exit data needs the operators to cooperate. A freeze is the operators declining to cooperate. So the same action that stops your payments stops you collecting the data you would need to leave. If you already hold a current copy, you can exit while frozen and no one can take that copy back. If you do not, you cannot obtain it afterwards, because the door you would fetch it through is the door that just closed. Self-custody on Spark is real, and it is conditional on having done that work before a freeze rather than after.
Three details make the condition harder than it first sounds. Your leaf data changes with every payment, so a backup from last week quietly omits the coins you received since. A targeted freeze, according to the review at spark.exposed, can look identical to an ordinary connection error, so you may not know you are being frozen until the moment to act has passed. And that same review checked Primal, Radar and Wallet of Satoshi and found their self-custody claims were not matched by a working exit path a normal user could run alone. Blink had to build that tooling themselves to show the exit was possible at all.
One assumption sits under everything else. The operators are meant to delete the keys they held with previous owners, but that deletion is a promise backed by reputation and legal agreements, not by mathematics. The Bitcoin Manual, written by a statechain developer, notes that operators who plan ahead can block a user's exit, and that the risk does not go away over time. A court order to retain keys and freeze an account is the kind of thing planned ahead.
What this means for a business.
None of this makes Spark dishonest. But conditional self-custody is a different thing from the permissionless kind, and the gap between them is the whole point. A prepared holder genuinely cannot have their Bitcoin seized, and the operators cannot move funds on their own. That is more than a custodial app offers, and Lightspark chose a slow, safe exit over a fast, fragile one on purpose. The point is narrower than a warning. On the main chain, your seed and the public ledger are always enough to recover your Bitcoin. On Spark, your seed plus a reachable company is the everyday case, and the day that company is not reachable is the day your seed alone may not be enough.
So return to the opening question. If the operators went offline tomorrow, could you get your Bitcoin out? The answer is not a plain yes or no. It is yes, if you hold current exit data, keep some Bitcoin on-chain to pay the fees, and run a wallet that can build and broadcast the exit without asking anyone. The test worth running is that simple: can your wallet complete an exit with the operators unreachable? A provider who can show you that has earned the word self-custody, and knowing where the permission ends is what lets a business decide how much to keep there.
