What Does It Cost to Force-Close a Lightning Channel? Evidence from 974,000 Channel Closes, 2018-2026

arXiv:2611.16311 · cs.CR, cs.NI · Submitted 2026-11-12 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.

Nadia: I'm Nadia.

Elias: And I'm Elias, here with Priya.

Nadia: We're looking at the paper: "What Does It Cost to Force-Close a Lightning Channel? Evidence from 974,000 Channel Closes, 2018-2026".

Elias: Here's the gist of the paper. Bitcoin’s Lightning payment network returns the funds of a channel that cannot close cooperatively through presigned transactions whose fees and delays fall on its operators The fee design of these transactions is changing now This paper measures the old design, across the fee regimes it lived through, before the migration, as a baseline for judging the new one The Core Findings on Force Close Costs Force closes have been counted and predicted The main answer is that fee spikes do not bring a wave of force closes A force close costs 2–6 times a cooperative close in fees and locks the closer’s balance for days, which outweighs the fees only in low-fee years The wait outweighs the fees only when fees are low When fees are high, pending payments are exposed Analysis of Fee Regimes and Spikes Fee spikes did not bring a wave of force closes Force closes in a spike week and the next were 0.97 times the four weeks before, against 1.02 after ordinary weeks of the same years Fixed commitment fees overpaid in calm markets The elasticity of the weekly force-close rate to the median fee rate of the same week is 0.00, at percentile 55 of a placebo that moves the fee series in time The Role of Anchor Outputs and Third Parties Anchor outputs became thirdparty dust Owners spent 81% of anchors in 2022 and 5% in 2026; third parties sweep the rest, from 2025 at the first block they can, and fees took 87% of the swept value in 2026 No P2WSH anchor was left unspent, while no third party has swept a simple-taproot anchor, so the premise of an open LND proposal to stop sweeping anchors holds only for the older kind Third-party sweeps are collected chain-wide, including anchors of commitments we never saw, and each sweep’s 330 sat per anchor is split between the fee (to miners) and the outputs (to the sweeper) The Impact of New Zero-Fee Commitments Zero-fee commitments would save only 12% of the anchor mechanism’s block space The first zero-fee commitment closed on 29 October 2025, before the specification was merged Across the anchor force closes we observe, the new format would have saved 12% of the anchor mechanism’s block space The measure of the migration is whether payments still resolve before their deadlines when the next spike comes Lifecycle Costs and Exposure A channel’s life costs a few thousand satoshis The median life cost 0.15% of the channel’s capacity In fees alone a force close is cheap: the median public force close cost 925 sat in 2026 The wait outweighs the fees only when fees are low The exposure, the HTLCs that must confirm before their deadlines while fees are high, has rarely become a loss Client Pair Dynamics and Channel Trends Among client pairs with more than 1,000 closes since May 2023, the force share ranges from 28% (LND–LND) to 40% (CLN–CLN) The force share of public closes fell from 44% in 2023 to 20% in 2026 Most of force closes now happen outside the public network In August 2026 the rate reached 51 per 1,000 open channels, and this month brought the Core Lightning shutdown and the largest contraction of the public graph Fee spikes do not break channels for pre-anchor or anchor channels The anchor design reduced that exposure for HTLCs, while its P2WSH anchors became dust that third parties sweep and miners collect Zero-fee commitments will save little block space The measure of the migration is whether payments still resolve before their deadlines when the next spike comes The cost of a force close is a ratio, a multiple of a cooperative close in fees and days of locked funds, and a wait, days before the closer’s own funds return Appendix on Force Share by Client Pair Among pairs with both ends labelled and more than 1,000 closes, the force share ranges from 28% (LND–LND) to 40% (CLN–CLN) The one outlier, LDK–LDK at 1.5% of 13,415 closes, is one node closing channels cooperatively at scale The chain does not show which side broadcast a force close, so a close is attributed to the pair The paper reports no new vulnerability: replacement cycling was disclosed in 2023

6: , and the splicing bug was fixed before the wave

31: All inputs are public blocks and public gossip The released aggregates name no node, address or transaction No list, ranking or label of nodes, sweeping operators, addresses or transactions, in the paper or in the released data The code is available at https://github.com/bitcoindatalabs/lnforce-close-costs All stages truncate their inputs at the data cutoff, block 974,041 (31 October 2026) The released aggregates name no node, address or transaction The underlying blocks are public; our channel list and the per-transaction tables are available from the author on request All closes to 4 October 2026 (four weeks for outputs to be spent) The open is a large part of that Closed Channels Force % Open Coop.

Nadia: Force All Range p90 % cap.

Elias: Opens % BTC = 1.29 36.3 8.00 Closed Channels Force BTC = 723 Total BTC % of fees % of block space = 0.15% 39.7 The median lifecycle if every output, change included, pays for its own size, and with the equal split is 3,435 Closed Channels Force life = 6,645 Cooperative life = 1970 Open alone = 321 Appendix on What a Public Channel’s Life Costs On-Chain Closed Channels Force % Open Coop.

Nadia: Force All Range p90 % cap.

Paper summary: Nadia: So to summarize this paper's thesis is that we need to measure the old fee design across all those different fee regimes before we move on to something new. The core finding they want you to hear is that fee spikes don't bring a wave of force closes, and the cost of a force close isn't what you think it is.

Elias: They claim a force close costs two to six times a cooperative close in fees and locks the closer’s balance for days, which only outweighs those fees when fees are low. They also look at how anchor outputs became third party dust and how zero-fee commitments would save very little block space, only twelve percent of it.

Priya: What matters for us is that they show that the exposure to late HTLC timeouts has rarely become a loss when fees are high, because the anchor design reduced that exposure for those transactions.

Nadia: They also track client pair dynamics, seeing force shares range from twenty-eight percent between LND and LND to forty percent between CLN and CLN for channels with more than one thousand closes since May two thousand twenty-three <ref:2611.16311#pg1>.

Elias: And they show a shift in behavior where the force share of public closes fell from forty-four percent in two thousand twenty-three down to twenty percent in two thousand twenty-six <ref:2611.16311#pg1>.

Priya: It seems like the paper is arguing that the old design, especially with its anchor mechanism, reduced that exposure for HTLCs, and third party sweeps are now collecting chain-wide, including anchors of commitments we never saw.

Nadia: So what this means for the world is they are measuring whether payments still resolve before their deadlines when the next spike comes against the old design structure.

Elias: It’s about understanding if the migration to zero-fee commitments actually solves the problem of force closes, or if it just changes where those costs land on chain.

Conclusion: Nadia: So looking at the full paper on "What Does It Cost to Force-Close a Lightning Channel? Evidence from nine hundred seventy-four thousand Channel Closes, two thousand eighteen-two thousand twenty-six" we see they've really been digging into the mechanics of this old system <ref:2611.16311#pg1,Evidence from 974,000 Channel Closes, 2018-2026>. They traced every public channel close and force close onchain from January two thousand eighteen to October two thousand twenty-six <ref:2611.16311#pg1,force close onchain from January 2018 to October 2026>.

Elias: The authors are essentially saying that the cost of a force close is a ratio, a multiple of a cooperative close in fees and days of locked funds, and then there's the wait time before the closer’s own funds return.

Priya: What this implies for us is that fee spikes don't break channels for those whose fee is fixed at signing or those who can bump it, because the anchor design actually reduced exposure to late HTLC timeouts when fees were high.

Nadia: They also found that zero-fee commitments will save little block space, maybe only twelve percent of the anchor mechanism’s block space.

Elias: The real measure of this migration is whether payments still resolve before their deadlines when the next fee spike comes, which is exactly what they are testing by comparing old and new formats.

Priya: It suggests that for those who can manage their fees, the old system had some built-in protection against massive losses from late confirmations during high fee periods.

Nadia: So it boils down to this: the cost of a force close is measured by a ratio of fees and locked funds multiplied by a wait time, and they’re checking if that ratio still holds up under the new zero-fee commitment rules.

Saurabh Kumar

Bitcoin Data Labs

cs.CR, cs.NI

Submitted: 2026-11-12

Updated: 2026-11-12

Comments: 14 pages, 8 figures, 8 tables. Code and aggregate data: https://github.com/bitcoindatalabs/ln-force-close-costs

Code: https://github.com/bitcoindatalabs/lnforce-close-costs

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 88/100

The gist: Bitcoin’s Lightning payment network returns the funds of a channel that cannot close cooperatively through presigned transactions whose fees and delays fall on its operators The fee design of these…

Key concepts

Force Close Costs
This measures the financial impact when a channel cannot close cooperatively. It is calculated as a ratio of fees and locked funds compared to a cooperative close. Force closes are cheap in terms of fees but involve locking funds for days, making the total cost dependent on when fees are high or low.
Fee Regimes
The paper analyzes how commitment fees have changed over time, noting that pre-anchor commitments paid significantly more than market rates during low-fee periods. The analysis shows that fixed commitment fees often overpaid in calm markets, influencing the overall cost structure of force closes.
Anchor Outputs
These are Bitcoin outputs used for anchoring commitments to the chain. The paper found that owners spent most anchors early on, and third parties sweep the rest. This mechanism has reduced exposure for HTLCs but creates a system where unspent anchors are collected by third parties.

Terminology

Summary

Bitcoin’s Lightning payment network returns the funds of a channel that cannot close cooperatively through presigned transactions whose fees and delays fall on its operators The fee design of these transactions is changing now This paper measures the old design, across the fee regimes it lived through, before the migration, as a baseline for judging the new one

The Core Findings on Force Close Costs

Force closes have been counted and predicted The main answer is that fee spikes do not bring a wave of force closes A force close costs 2–6 times a cooperative close in fees and locks the closer’s balance for days, which outweighs the fees only in low-fee years The wait outweighs the fees only when fees are low When fees are high, pending payments are exposed

Analysis of Fee Regimes and Spikes

Fee spikes did not bring a wave of force closes Force closes in a spike week and the next were 0.97 times the four weeks before, against 1. p. 002 after ordinary weeks of the same years Fixed commitment fees overpaid in calm markets The elasticity of the weekly force-close rate to the median fee rate of the same week is 0. p. 100, at percentile 55 of a placebo that moves the fee series in time

The Role of Anchor Outputs and Third Parties

Anchor outputs became thirdparty dust Owners spent 81% of anchors in 2022 and 5% in 2026; third parties sweep the rest, from 2025 at the first block they can, and fees took 87% of the swept value in 2026 No P2WSH anchor was left unspent, while no third party has swept a simple-taproot anchor, so the premise of an open LND proposal to stop sweeping anchors holds only for the older kind Third-party sweeps are collected chain-wide, including anchors of commitments we never saw, and each sweep’s 330 sat per anchor is split between the fee (to miners) and the outputs (to the sweeper)

The Impact of New Zero-Fee Commitments

Zero-fee commitments would save only 12% of the anchor mechanism’s block space The first zero-fee commitment closed on 29 October 2025, before the specification was merged Across the anchor force closes we observe, the new format would have saved 12% of the anchor mechanism’s block space The measure of the migration is whether payments still resolve before their deadlines when the next spike comes

Lifecycle Costs and Exposure

A channel’s life costs a few thousand satoshis The median life cost 0.15% of the channel’s capacity In fees alone a force close is cheap: the median public force close cost 925 sat in 2026 The wait outweighs the fees only when fees are low The exposure, the HTLCs that must confirm before their deadlines while fees are high, has rarely become a loss

Client Pair Dynamics and Channel Trends

Among client pairs with more than 1,000 closes since May 2023, the force share ranges from 28% (LND–LND) to 40% (CLN–CLN) The force share of public closes fell from 44% in 2023 to 20% in 2026 Most of force closes now happen outside the public network In August 2026 the rate reached 51 per 1,000 open channels, and this month brought the Core Lightning shutdown and the largest contraction of the public graph

Conclusion on Design Trade-offs

Fee spikes do not break channels for pre-anchor or anchor channels The anchor design reduced that exposure for HTLCs, while its P2WSH anchors became dust that third parties sweep and miners collect Zero-fee commitments will save little block space The measure of the migration is whether payments still resolve before their deadlines when the next spike comes The cost of a force close is a ratio, a multiple of a cooperative close in fees and days of locked funds, and a wait, days before the closer’s own funds return

Appendix on Force Share by Client Pair

Among pairs with both ends labelled and more than 1,000 closes, the force share ranges from 28% (LND–LND) to 40% (CLN–CLN) The one outlier, LDK–LDK at 1.5% of 13,415 closes, is one node closing channels cooperatively at scale The chain does not show which side broadcast a force close, so a close is attributed to the pair The paper reports no new vulnerability: replacement cycling was disclosed in 2023 [6], and the splicing bug was fixed before the wave [31] All inputs are public blocks and public gossip The released aggregates name no node, address or transaction No list, ranking or label of nodes, sweeping operators, addresses or transactions, in the paper or in the released data The code is available at https://github.com/bitcoindatalabs/lnforce-close-costs All stages truncate their inputs at the data cutoff, block 974,041 (31 October 2026) The released aggregates name no node, address or transaction The underlying blocks are public; our channel list and the per-transaction tables are available from the author on request All closes to 4 October 2026 (four weeks for outputs to be spent) The open is a large part of that Closed Channels Force % Open Coop. p. 8 Force All Range p90 % cap. p. 8 p. 8 Opens % BTC = 1. p. 729 36.3 8.00 Closed Channels Force BTC = 723 Total BTC % of fees % of block space = 0. p. 315% 39.7 The median lifecycle if every output, change included, pays for its own size, and with the equal split is 3,435 Closed Channels Force life = 6,645 Cooperative life = 1970 Open alone = 321

Appendix on What a Public Channel’s Life Costs On-Chain

Closed Channels Force % Open Coop. Force All Range p90 % cap. Opens % BTC Range p90 % cap. p. 8 Opens % BTC = 6,645 3,435 2,804–3,435 Closed Channels Force life = 6,645 Cooperative life = 1970 Open alone = 321

Appendix on What Anchor Outputs Are Used For

Year Anchors Owner % Third % Unspent % Bumped % Delay Sweep txs Per tx sat/vB Miners % Break-even fee rate = 6.38 sat/vB The break-even fee rate is 6. p. 338 sat/vB The median third-party sweep of a public channel’s anchor came 1,708 blocks after a 2023 close, about 12 days, against exactly 16 from 2025 Third-party sweeps are collected chain-wide, including anchors of commitments we never saw The missing key does not explain the unspent anchors: for 14,828 of the 22,158 unspent anchors the key has been on-chain for a median of 101 days The unswept taproot anchors are worth little to a sweeper The owners’ own spends stopped as well, even though adding the anchor to a sweep still paid

Appendix on Fee Regimes

Pre-anchor commitments paid twice the market when fees were low Anchor commitments paid about the market Pre-anchor commitments paid a median 1.1–1.3 times the median fee rate of their own block from 2018 to 2024, and 2. p. 30 times in 2025–26 Anchor commitments paid a median 0. p. 096–1.21 times their block’s median The commitment’s fee rate divided by the median fee rate of the block it confirmed in, by year: median and interquartile range, pre-anchor and anchor commitments

Appendix on Zero-Fee Commitments

Across the 223,587 anchor force closes we observe, the anchor mechanism took 173 vB per close Zero-fee commitments would have taken 152 vB That is 12% less The gains lie elsewhere: the 1.28 BTC held in 330-sat anchor outputs would not exist, commitment fees no longer have to be guessed at signing, and package relay replaces the carve-out that Bitcoin Core has removed The test is the one on which anchors improved on the old format: whether HTLC transactions confirm before their deadlines when fees spike

Appendix on What a Force Close Costs

In fees alone a force close is cheap: the median public force close cost 925 sat in 2026 The wait outweighs the fees only when fees are low The break-even rate was 2.7% in 2025 and 1. p. 02% in 2026, so in the low-fee years even a modest return on the locked funds made the wait the larger cost In fees, a force close is cheap and a channel’s life costs a fraction of a percent of its capacity The cost of a force close is a ratio, a multiple of a cooperative close in fees and block space, and a wait, days before the closer’s own funds return

Appendix on HTLC Deadlines

Of 70,808 HTLC-timeout transactions, the median confirmed 1–3 blocks after its expiry in every year When the block median fee rate at expiry was 50 sat/vB or more, 30.5% of pre-anchor HTLC timeouts confirmed more than 40 blocks late Anchors keep HTLC timeouts closer to the deadline when fees are high

Appendix on Fee Spikes

Spikes still matter, but through what happens after a close The caveat is that the chain records confirmation, not broadcast Unlisted force closes fell when fees rose (elasticity −0.22) Fee spikes do not bring waves of Lightning force closes, for channels whose fee is fixed at signing or for those that can bump it The measure of the migration is whether payments still resolve before their deadlines when the next spike comes

In fees alone a force close is cheap and a channel’s life costs a fraction of a percent of its capacity The cost of a force close is a ratio, a multiple of a cooperative close in fees and block space, and a wait, days before the closer’s own funds return The exposure, the HTLCs that must confirm before their deadlines while fees are high, has rarely become a loss The anchor design reduced that exposure for HTLCs

Appendix on What the Chain Can Catch

Two 2026 events show what transactions alone can reveal A week of revoked commitments matched a splicing bug that Core Lightning had fixed days before, and the only split resolutions in the order replacement cycling leaves were a test of that attack on mainnet We found both without node logs, and the same tests can be run on new blocks as they arrive The chain does not show what made each counterparty close at the lock, or whether the funder caused it

Appendix on Limitations

The chain does not show which party broadcast a force close, so costs are per channel, both parties together Invisible bumps. Fee bumps that never confirmed, and replaced transactions, leave no trace on the chain; we have no mempool record Private channels. p. 2 Private channels appear only when they force close, so their number and their force-close rate are unknown Open channels. p. 2 Channels whose nodes have left count as open until closed on-chain, which lowers the per-channel rate as they accumulate Penalties. p. 2 Only punished breaches are visible The scope of the cost account is limited by design The breakeven rate counts the broadcaster’s locked balance only, not pending HTLC amounts Simple taproot channels’ spends are not classified, so their delays, HTLCs and bumps are missing from §VI; they were 23% of 2026’s unlisted force closes The fee regimes depend on the fee regime, which is why the window covers all of Lightning’s history The association is a within-year association; it rules out a large same-week effect, not every causal path

Appendix on Ethics

All inputs are public blocks and public gossip We report aggregates only: no list, ranking or label of nodes, sweeping operators, addresses or transactions, in the paper or in the released data In particular we do not identify the dominant anchor sweeper, the nodes of the first zero-fee commitment closes The paper reports no new vulnerability: replacement cycling was disclosed in 2023 [6], and the splicing bug was fixed before the wave [31]

Appendix on Conclusion

Fee spikes do not bring waves of Lightning force closes, for channels whose fee is fixed at signing or for those that can bump it What a force close costs is a multiple of a cooperative close in fees and days of locked funds, and the wait outweighs the fees only in low-fee years When fees are high, pending payments are exposed, yet forwarders lost few of them, each with the preimage already on-chain The anchor design reduced that exposure for HTLCs Zero-fee commitments will save little block space The measure of the migration is whether payments still resolve before their deadlines when the next spike comes

Appendix on Data and Code Availability

Code, the aggregate datasets behind every figure and table, and a script that recomputes every number quoted in the text are available at https://github.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Adaptive Fee Regime Modeling: Improve AI systems to predict channel closure risk by integrating fee spike data with historical patterns, specifically by using Fee timing b Null 95% (C) Pctl. to determine if a close is likely to occur during a fee surge.

  2. Cost-Based Risk Assessment: Enhance the ability of AI systems to quantify the true cost of forcing a close by calculating a ratio against cooperative closes, considering fees, days locked, and block space usage across different commitment formats (pre-anchor vs. anchor).

  3. Anchor Output Behavior Prediction: Develop models that predict whether P2WSH anchor outputs will be used for fee bumping or cleanup based on historical data showing Bumped % Anchors and the behavior of third-party sweeps versus owner spends.

  4. Zero-Fee Commitment Evaluation: Improve systems to assess the block space efficiency of new protocols by comparing Zero-fee commitments against the old mechanism, specifically noting that they would have saved 12% of the anchor mechanism’s block space.

  5. HTLC Exposure Analysis: Refine risk assessment for forwarding nodes by analyzing how often HTLC timeouts confirm after expiry, using data like pre-anchor HTLC timeouts confirmed more than 40 blocks late when fees are high.

Sources

Related papers