# Who Patches the Lightning Network? Evidence from Release Lines in Gossip, 2023-2026

Crypto.world

> Gossip feature bits expose node release lines, yielding the first measured evidence that Lightning patching follows app-store release schedules rather than security advisories.

**arXiv**: 2610.30350

**Published**: 2026-10-21

**Authors**: Saurabh Kumar

## AI summary

## Overview

**Research area:** Measurement of software security practices in a decentralised payment network — specifically, how quickly operators of Bitcoin Lightning Network nodes install patched releases, and who actually drives those upgrades.

**Technical level:** Intermediate. The paper is readable without deep protocol knowledge, but it assumes some familiarity with the Lightning Network, node announcements ("gossip"), and basic statistical survival analysis.

**Scope:** The paper reverse-engineers the software release line of public Lightning nodes from gossip feature bits, validates that inference, and then uses 814 daily snapshots of the public graph (May 2023 to 13 October 2026) to measure upgrade timing, platform-driven upgrade waves, and operator response to 30 security advisories affecting LND and Core Lightning.

## What This Paper Is About

Lightning nodes hold user funds in payment channels, so a software vulnerability can put money at risk — and throughout 2026 every major implementation (LND, Core Lightning, Eclair, LDK) published security advisories telling operators to upgrade. Nobody had ever measured whether operators comply, because Lightning gossip carries no version string, unlike Bitcoin and Ethereum handshakes. The paper shows that a node's *release line* can nevertheless be inferred from the protocol feature bits it announces, then uses that inference to ask who patches the network: operators acting on their own, the node-in-a-box platforms that package the software, or the advisories themselves.

## Key Contributions

1. **Release-line inference from gossip.** The authors build a table of default, optional and removable feature bits from the source code of every LND (0.7–0.21.4), Core Lightning (0.8.2–26.06.9), Eclair (0.3–0.14.3) and LDK (0.0.113–0.2.7) release, group releases with identical defaults into "release lines" (LND has 10 lines from 0.8 to 0.21, CLN 13, Eclair 13, LDK 6), and match observed node feature sets against them. Validation on CLN nodes that state their version yields 132 of 136 node–line pairs exact (97%, 95% CI 93–99%).
2. **Upgrade dynamics.** Using interval-censored survival estimation on nodes live at each release, the paper quantifies how long operators take to move to a new line, and identifies which node characteristics predict faster upgrading, while excluding a managed fleet of up to 1,781 live LND nodes that would otherwise dominate per-node statistics.
3. **Platforms, not projects, set the pace.** By dating Umbrel app-store merges to the minute and BTCPay Server's security release 2.4.2 of 7 August 2026, the paper shows the largest LND upgrade bursts follow platform releases, not LND releases or advisories.
4. **Disclosure response and residual exposure.** For 30 advisories whose fix boundary is visible in gossip, the paper measures exposure at disclosure, tests whether exposed nodes leave affected lines faster afterwards, and tracks what became of the exposed cohorts.

## Main Findings

- **Version inference is nearly exact where ground truth exists.** Across 15,549 announced node-days of 124 CLN nodes on 11 lines, the "strict range" contains the stated line on 99.99% of node-days and the "likely line" equals it on 98.2%.
- **Three independent checks support the inference for LND, which states no version.** Line decreases are 0.7% of observed line changes (against 13,802 increases); no LND line released since 2024 was seen on more than 70 nodes before its publication, versus 2,628–3,756 in its first 90 days; and 97–99.7% of the 1,165 line changes signed within 48 hours of one of four platform releases landed on the line that release shipped.
- **Half of LND nodes upgrade within 88 days; the tail is small or already behind.** 76% upgrade within a year. CLN's median is 137 days. The smallest quarter of LND nodes by live capacity took a median 135 days, the largest quarter 75. A node more than one line behind at release has an odds ratio of 0.19 for moving in any given week; Tor-only nodes move faster (1.61).
- **The three largest LND upgrade waves were Umbrel's.** Each wave (262–385 nodes) followed Umbrel publishing a new LND version 19–41 days after LND's own release, with at least 87% of nodes signing within 48 hours of the merge (median 16–27 hours after) and at most 1% in the 24 hours before. A fourth wave of 114 nodes on 8 August 2026 followed BTCPay Server's own security release.
- **Platform windows are short but carry a disproportionate share of upgrades.** Since June 2025, the 48 hours after a platform release cover 4.4% of the time but hold 17% of LND upgrades (CLN: 14% against 5.4%). These are lower bounds, since several platforms could not be dated.
- **Fixes ship long before disclosure.** The median gap between first fixed release and public disclosure is 256 days; six advisories Lightning Labs published on 11 August 2026 concerned bugs fixed in LND 0.19.0, released 446 days earlier.
- **Disclosure barely moved LND operators.** Across nine LND disclosures, rate ratios after versus before disclosure are at most 1.48 and none is significant (permutation p ≥ 0.12). Two months after the August 2026 batch, 71% of the nodes exposed at disclosure were still live on an affected line; only 10% had upgraded and 20% were no longer live.
- **One CLN disclosure did move operators — the ones already close to current.** CLN's 19 July 2026 gossipd advisory raised the weekly exit rate from affected lines from 1.6% to 6.2%; outside platform moves the ratio is 3.28 (p = 0.003), the only one of twelve tests below a Bonferroni threshold of 0.05/12. Among the 334 nodes exactly one line behind, the ratio was 3.87 (p = 0.003).
- **Nodes on older lines moved when their platform told them to, not when CLN did.** Their outside-platform ratio is 1.54 (p = 0.212); most of their movement came in the 48 hours after BTCPay's 7 August security release.
- **The "restart offline" call reached mostly already-patched nodes.** After CLN's 26 August 2026 call, 23.4% of the 334 live nodes on the newest line were no longer live two days later, against 11.0% of the 310 on older lines (odds ratio 2.5, Fisher p = 4×10⁻⁵) — against medians of 3.6% and 3.1% in the 21 preceding two-day August windows.
- **Counting announced nodes instead of live ones badly distorts exposure.** On the cutoff day, 2,689 LND nodes announced a release older than 0.19, but only 434 were live: 42.3% of announced nodes versus 12.5% of live nodes.

## Methodology in Plain English

The authors run their own LND node and dump the public graph once a day, keeping every node announcement (feature bits, alias, colour, addresses, signature time) and every channel policy. A quality filter discards partial, stale or wallet-locked dumps, leaving 814 days of usable snapshots out of the May 2023 – October 2026 window.

They then read the source code at every release tag of the four implementations and record which feature bits each release sets by default, which an operator can switch on, and which can be switched off. Releases with identical defaults are collapsed into a "release line," the finest distinction gossip can make — patch releases are invisible. A node's observed set of bits is compared against each line: a line is "consistent" if the node's bits could be produced by that line under some non-default configuration. The "strict range" is every consistent line released by the observation day; the "likely line" is the consistent line needing the fewest non-default bits, dated from the day the node's current bit set first appeared (so a new release cannot retroactively relabel existing nodes).

For upgrades, they follow each node that was live on an older line at a release day and record when it is first seen on the new line or later. Because snapshot gaps make the exact upgrade day uncertain, they use a nonparametric maximum-likelihood estimator for interval-censored data, plus a discrete-time logit model with week and release fixed effects and standard errors clustered by node.

For platforms, they take the merge time of every Umbrel app-store pull request that changed the LND or CLN version, and BTCPay Server's security release, and compare those timestamps against the signature times of new node announcements. For disclosure response, they count weekly moves off wholly affected lines before and after each advisory and use quasi-Poisson intervals (weekly counts vary up to 70 times more than Poisson) plus exact permutation tests; they also recompute everything excluding moves signed within 48 hours of a platform release, to separate chosen upgrades from pushed ones.

"Liveness" is read from peers rather than from the node itself: a node is live on a day if at least one channel direction towards it, set by a peer, is enabled and was updated in the previous 14 days. A managed fleet of `LNT.<name>` nodes is excluded from all per-node statistics.

## Why This Matters

**Impact on research.** Longitudinal Lightning studies have measured topology, centralisation, channel lifecycles and node lifetimes, but never software versions, because gossip was assumed to carry no version. This paper establishes that release lines are recoverable and validated, opening a measurement channel that did not previously exist — and it ports the two-decade patch-deployment literature (Arbaugh, Rescorla, Durumeric, Sarabi) to a network where fixes often precede disclosure by months.

**Real-world applications:**

- **Maintainers deciding where to spend effort.** The paper's own conclusion is blunt: the lever on lagging nodes is the platforms that package releases, not the advisories. Coordinating with Umbrel and BTCPay reaches far more stragglers than a mailing list post.
- **Node-in-a-box and merchant-stack operators.** Platform releases account for 17% of LND upgrades in 4.4% of the time, so a platform's shipping cadence is effectively a security control for its users.
- **Risk assessment for channel counterparties and routing nodes.** Knowing the share of live capacity on affected lines at disclosure (for example 52.5% for the cooperative-close disclosure) gives a quantified view of how much value sits behind unpatched code.
- **Disclosure policy design.** The 256-day median gap between fix and disclosure, and the near-total absence of LND response, are direct evidence for how coordinated disclosure plays out when patches are silently available for months.

**Industry relevance.** The paper quantifies a fundamental structural difference between Lightning and other blockchain networks: Bitcoin and Ethereum peers report versions in the handshake and can be counted directly, and Zcash expires old releases at network upgrades, but Lightning has neither a version string nor forced expiry. The 71% residual exposure two months after a ten-advisory batch is an argument for changing that.

## Future Directions

- **Extending version visibility beyond feature bits.** Patch releases are invisible and developer overrides can imitate any feature set; a gossip-level version field, or platform-attested build metadata, would close the gaps the authors can only bound.
- **Covering the platforms that could not be dated.** Start9, RaspiBlitz, myNode and hosted services are excluded, so the 17% and 14% platform-window shares are stated lower bounds; dating them would tighten the central claim.
- **Reaching unannounced nodes.** Mobile wallets and services behind unannounced channels — where much of LDK and Eclair runs — are invisible to gossip, and Eclair had too few live public nodes (about ten) for meaningful shares, leaving two of the four implementations essentially unmeasured.
- **Testing whether platform-mediated delivery actually shortens the window of vulnerability.** The paper shows platforms cause upgrades but does not show that this reduces exploitation; CLN's maintainers reported attacks on unpatched nodes by October 2026, which is the outcome variable a follow-up would need.

## Target Audience

Lightning implementation maintainers and security teams; operators of node-in-a-box platforms and merchant stacks such as Umbrel and BTCPay Server; blockchain measurement researchers interested in software-update behaviour; and anyone working on vulnerability disclosure policy for decentralised systems. Readers wanting the raw numbers will find them in the paper's tables (I–V) and figures (1–7); the authors state that all data and code are released, with aggregates only and no per-node labels.

## Authors’ abstract

Every implementation of Bitcoin's Lightning payment network disclosed security vulnerabilities in 2026, and every advisory told operators to upgrade. Whether they do has never been measured, because Lightning gossip carries no software version. Matching the feature bits a node announces against the source of every LND, Core Lightning, Eclair and LDK release, we show that gossip reveals a node's release line after all. Where Core Lightning nodes state their version, the…

[The full abstract on arXiv](https://arxiv.org/abs/2610.30350)

[Read the original paper](https://arxiv.org/abs/2610.30350)

---

Canonical: https://crypto.world/paper/who-patches-the-lightning-network-evidence-from-release-lines-in-gossip-2610.30350
