mirror of https://github.com/luxfi/lattice-analysis.git synced 2026-10-30 16:21:35 +00:00

Files

T

zeekayandHanzo Dev dbc56f1364 chore: initial commit — tree had no git at all

Lattice analysis rig. No history, no remote, not on GitHub.

Excludes three compiled analysis binaries (lattice_test, round2,
kurtosis_sweep — 6.1 MB each) that are build output, not source; committing
them was what made the initial push fail.

Co-authored-by: Hanzo Dev <[email protected]>

2026-07-28 11:57:48 -07:00

3.3 KiB

Raw Permalink Blame History

LUX Lattice Security Analysis Rig

Dragonfire / In2Infinity Research

Self-contained analysis toolset for the LUX lattice cryptography stack. Instruments the real luxfi/lattice library — no mocks, no placeholders.

Quick Start

# Prerequisites: Go 1.25+, Python 3 with numpy/scipy/matplotlib

# 1. Clone the LUX repos (if not already done)
cd ~/projects/lux-review
git clone https://github.com/luxfi/lattice.git
git clone https://github.com/luxfi/consensus.git

# 2. Set up the rig
cd rig/
go mod tidy

# 3. Run the full test battery
./run_all.sh

What It Tests

Tool	Language	What It Does
`lattice_analysis.go`	Go	Captures Gaussian noise, applies NTT, measures timing (side-channel detection)
`kurtosis_sweep.go`	Go	Parameter sweep: kurtosis across N=256-4096, multiple moduli, 20 trials each
`round2_tests.go`	Go	Full test battery: KS tests, sigma/bound/modulus variation, DFT vs NTT
`spectral_analysis.py`	Python	PSD, autocorrelation, distribution charts, HTML report
`phi_analysis.py`	Python	Brillouin zone geometric ratio analysis
`round2_analysis.py`	Python	KS/chi-square tests, constant discrimination, summary charts

Key Findings (Confirmed)

CRITICAL: Gaussian sampler timing side-channel (CV=0.376)

lattice/ring/sampler_gaussian.go:138 — rejection sampling creates data-dependent timing
Mean: 42,700ns, range: 31,178ns to 808,183ns (26x spread)
NIST dropped Gaussian sampling from Kyber/Dilithium for exactly this reason

HIGH: NTT transforms Gaussian to uniform

Confirmed across 520+ trials, all parameters
Excess kurtosis = -1.197 (uniform = -1.200)
KS test confirms: p > 0.05 for uniformity, p ~ 0 for Gaussianity
This is specific to cyclotomic NTT (plain DFT does not do this)
Mathematically expected (smoothing lemma) — not a vulnerability, but validates methodology

HIGH: bytes.Equal in witness verification

consensus/protocol/quasar/witness.go:233 — timing oracle on proof comparison

Running Individual Tests

# Timing + noise analysis (produces results/ directory)
go run lattice_analysis.go

# Parameter sweep across ring degrees
go run kurtosis_sweep.go

# Full A-E test battery
go run round2_tests.go

# Python analysis (run after Go harnesses)
python3 spectral_analysis.py      # Charts + HTML report
python3 phi_analysis.py           # Brillouin zone analysis
python3 round2_analysis.py        # KS tests + constant discrimination

Output

All results go to results/:

summary.json — timing + spectral analysis summary
kurtosis_sweep.json — parameter sweep data
round2_results.json — full test battery results
phi_analysis.json — geometric constant analysis
report.html — visual report with charts
charts/ — PNG charts for all analyses
*.csv — raw data for external analysis

Extending

To test with different parameters, edit the constants at the top of each Go file:

LogN — ring degree (8=256, 10=1024, 12=4096)
Sigma / Bound — Gaussian distribution parameters
Moduli — NTT-friendly primes (must satisfy q ≡ 1 mod 2N)
NumTimingTrials / Trials — number of measurements

Requirements

Go 1.25+ (tested with 1.25.6)
Python 3.10+ with: numpy, scipy, matplotlib
~500MB disk for repos + results
~2 minutes for full test battery

3.3 KiB Raw Permalink Blame History