Code, aggregate data and LaTeX source for 'Who Patches the Lightning Network? Evidence from Release Lines in Gossip, 2023-2026' (arXiv:2610.30350)
Saurabh Kumar, Bitcoin Data Labs
Paper: arXiv:2610.30350 (HTML, PDF) · DOI 10.48550/arXiv.2610.30350 · Bitcoin Data Labs research. Builds on Does Client Diversity Protect the Lightning Network? (arXiv:2610.20952), which owns the client-inference rules used here.
Code, aggregate data and LaTeX source for the paper. arXiv categories: cs.CR, cross-list cs.NI.
Data cutoff: 2026-10-13.
Lightning gossip carries no software version, so whether operators act on security advisories has never been measured. The paper infers the release line of every public LND, Core Lightning, Eclair and LDK node from the feature bits it announces, matched against the source of every release, and applies it to daily gossip snapshots from May 2023 to October 2026. The network patches on release and platform schedules, not on disclosures: half of the LND nodes on an older line upgrade within 88 days of a release; the largest upgrade waves came about a day after the Umbrel app store shipped a new LND version; and of 12 disclosures that can be tested, only Core Lightning's of July 2026 raised the rate at which nodes left affected releases, and only among nodes one release line behind; Core Lightning's call to restart offline was followed mostly by nodes already on the newest line. Two months after Lightning Labs' August 2026 batch, 71% of the nodes exposed at disclosure still ran an affected release.
Can gossip tell which software release a Lightning node runs? Its release line, yes. Matching the feature bits a node announces against the source of every LND, Core Lightning, Eclair and LDK release reveals it; where Core Lightning nodes state their version, the inferred line is exact for 132 of 136 node-line pairs.
How fast do Lightning nodes upgrade? Half of LND nodes upgrade within 88 days of a release. Small nodes and those that skipped releases lag far behind.
Do operators upgrade on security advisories or on platform releases? The network patches on release and platform schedules, not on disclosures. The three largest upgrade waves came about a day after the Umbrel app store shipped a new LND version, and since June 2025, 17% of LND upgrades came in the 48 hours after a platform release, windows that cover 4.4% of the time. Of 12 disclosures we can test, only Core Lightning's of July 2026 sped up upgrades, and only among nodes one release line behind.
How many nodes stay exposed after a disclosure? Two months after Lightning Labs' August 2026 batch, 71% of the LND nodes exposed at disclosure still ran an affected release.
paper/: LaTeX source (main.tex, sections/, references.bib, figures/, tables/, numbers.tex), the
compiled main.pdf and a build script.src/01_*.py … src/05_*.py: the five pipeline stages.src/lnmetrics/: the frozen inference rules and reference data: client rules (client_fingerprint.py), the
release table with the feature bits of every release (ln_release_features.json, read by
version_fingerprint.py), the advisory list with the public source of each disclosure (advisories.py), the dated
platform releases (platforms.py), and the gossip reader and panel helpers (gossip.py, panel.py).data/: aggregate JSON outputs behind every figure, table and in-text number.scripts/package_arxiv.py: builds the arXiv upload bundle.pip install -r requirements.txt
The pipeline reads daily snapshots of the public Lightning gossip graph as Parquet files; src/lnmetrics/gossip.py
documents the expected layout (gossip_daily/{nodes,policies,_quality.parquet}). The underlying gossip data are
public; our daily snapshots are available from the author on request. Point the pipeline at them with
export LN_PARQUET_DIR=/path/to/parquet # default: data/parquet
Each stage truncates its inputs at DATA_CUTOFF (src/config.py) and writes its figures to paper/figures/, its
tables to paper/tables/ and its numbers to data/*.json.
python src/01_build_panel.py # node-day panel: client, release line, liveness, fleet flag (~7 min) -> data/panel/
python src/02_validate_lines.py # release-line validation and consistency checks -> Fig. 1, Table I
python src/03_upgrade_dynamics.py # adoption, time to upgrade, hazard model, platform waves -> Figs. 2-5, Table II
python src/04_advisory_exposure.py # exposure per advisory, response to disclosure, robustness -> Figs. 6-7, Tables III-V
python src/05_supporting_numbers.py # every number quoted in the text -> paper/numbers.tex
The node-day panel (data/panel/, about 1.3 GB) is not committed.
paper/build.ps1 (Windows, MiKTeX) runs pdflatex → bibtex → pdflatex until references settle; build.bat runs it
from cmd. On other systems:
cd paper && pdflatex main && bibtex main && pdflatex main && pdflatex main
python scripts/package_arxiv.py then writes the arXiv bundle to dist/.
Aggregates only: no list, ranking or label of individual nodes by release, platform or exposure is published, in the
paper or in data/.
Please cite the paper (GitHub's Cite this repository button gives the same entry from CITATION.cff):
@misc{kumar2026upgradedynamics,
title = {Who Patches the Lightning Network? Evidence from Release Lines in Gossip, 2023--2026},
author = {Saurabh Kumar},
year = {2026},
eprint = {2610.30350},
archivePrefix = {arXiv},
primaryClass = {cs.CR},
doi = {10.48550/arXiv.2610.30350},
url = {https://arxiv.org/abs/2610.30350}
}
MIT (see LICENSE).