Skip to content
Crypto.world

The Ledger

Preprint proposes journal-ordered withdrawal permits for Bitcoin custody

A paper proposes a Bitcoin custody design that orders withdrawal permits against pauses and can revoke an individual signing lease. Its prototype was tested on regtest, synthetic schedules and HTTP workloads, with results and limitations reported by the authors.

Preprint proposes journal-ordered withdrawal permits for Bitcoin custody

A permit tied to a shared journal

An arXiv preprint proposes a Bitcoin custody design that makes withdrawal signing permits subject to an ordered journal, allowing a pause to block new permits and a chain reorganization to restore exposure from previously closed transactions. The paper, submitted to arXiv on Nov. 18, calls the prototype Release-Bound Withdrawal Authorization, or RBWA. Its authors are Jian Cui of the University of Illinois Urbana-Champaign, Rui Zhu of Yale University and Haixu Tang of Indiana University Bloomington.

The design addresses a scenario in which a compromised withdrawal backend asks a wallet signer to approve payments that customers did not authorize. It combines customer approval, balance evidence and transaction checks with a replicated journal that records pauses and permit issuance. The signer decodes proposed Bitcoin transactions and checks recipients, amounts, change output and fees against the authorization before issuing a permit bound to the transaction’s exact bytes.

Pause and permit issuance use the same compare-and-swap ordering mechanism. If the pause is recorded first, a concurrent attempt to issue a new permit fails; resuming increments an epoch so a reservation made before the pause cannot become newly eligible afterward. A permit already issued remains charged even if its response was lost or signing happens later, reflecting the possibility that signature material may already have escaped.

Reorganizations and large payouts

RBWA has two accounting modes for a transaction whose closing evidence is later removed from Bitcoin’s active chain. In the tracked mode, the signer checks the chain evidence and reopens the permit’s charge if its anchor is missing, placing the wallet lane on hold. The alternative retained mode never recycles the charge for an issued permit, even after settlement or a confirmed conflict, which the paper says imposes a finite cumulative issuance limit for that lane.

For an optional large-payout route, a wallet lane can stage a batch once its authorized principal reaches a configured threshold. The payout waits for a block-based relative delay, while a separately controlled recovery key can spend the staged output without that delay. The journal orders payout authorization against recovery requests, but the paper notes that an already-held payout can still compete once it matures. Smaller independently authorized batches can use the direct route.

A separate lease profile lets operators revoke a particular signing lease. A lease is tied to a wallet, lane, workload, holder key and validity interval, and the holder signs a message for a proposed transaction. Authenticated challenge receipts or audit evidence can trigger a journal fence; if that fence is recorded before release, it cancels the lease’s unissued reservations. Permits already issued remain charged, and a complete signature held elsewhere can still be broadcast under Bitcoin’s rules.

Prototype results and limits

The authors implemented the prototype with a three-member etcd cluster and tested Bitcoin transactions on regtest. Direct payments confirmed in one block; staged payments confirmed in three or seven blocks with delays of two or six blocks. Those are test results, not production-network performance figures.

In matched synthetic schedules, the paper reports that probes reduced permits through compromised leases from 20 to 9 out of 33, with one false fence. Two held-out HTTP workloads had equal fixed-threshold precision for boosted audit and probe-augmented detectors, and the authors say the audit fields already exposed the probe-authentication outcome. Live tests also checked a maintenance exception and a file marker; the marker fenced a compromised lease that an RPC-only audit missed.

The work is a prototype for a narrowly specified Bitcoin withdrawal profile. Its stated trust assumptions include an intact customer confirmation key, an independent ledger, a mandatory policy signer and a crash-fault-tolerant log. The paper also says that a chain can change after a sampled observation and that a transaction already fully signed can be submitted outside the prototype’s publisher; the design restores accounting when it detects a rollback but does not make the journal and Bitcoin chain atomic.

Sources

  1. Original source · Preprint arXiv arXiv:2611.25231

Related stories

Explore

More articles