ε-tx

Privacy analysis for cryptocurrency transactions. Computes information-theoretic bounds on what a blockchain observer can learn. Bitcoin, Monero, Lightning, cross-chain.

Run it

git clone https://github.com/Giuseppe552/epsilon-tx.git && cd epsilon-tx
npm install && npm run build

etx analyse bc1q...                                        # Bitcoin address
etx analyse bc1q... --adversary law-enforcement             # who's watching?
etx batch bc1q... bc1p... --timeline                        # full wallet privacy history
etx batch bc1q... --csv > privacy.csv                       # plot privacy over time
etx coinjoin abc123...                                      # Boltzmann entropy
etx classify abc123...                                      # Chainalysis classification
etx ring abc123... --optimise                                # Monero optimal decoys
etx route 02abc... 03def...                                  # Lightning privacy vs cost
etx crosschain 0xabc...                                     # bridge linking risk
etx network --node-type electrum --broadcast direct --rpc    # network leakage

Every command outputs JSON to stdout. Pipe into jq, Python, anything.

8 attack surfaces

#	Surface	Method	Reference
1	UTXO clustering	Co-spend graph + spectral sub-clustering (Fiedler vector)	Meiklejohn (2013), Fiedler (1973)
2	Wallet fingerprinting	8 heuristics x 7 profiles, log-likelihood scoring	Ishaana Misra (2024)
3	Amount analysis	Shannon entropy, roundness detection, cross-tx correlation	Androulaki (2013)
4	Timing + network	Circadian timezone, DFT periodicity, KS test, IP/RPC leakage	Biryukov (2014), CVE-2025-43968
5	CoinJoin	Boltzmann entropy, sub-tx decomposition, post-mix degradation	LaurentMT (2016), Maurer (2025)
6	Monero rings	Likelihood ratio analysis, inverse-OSPEAD optimal decoys	OSPEAD (2025), Moser (2018)
7	Lightning	Sender anonymity, timing leakage, balance probing, Pareto frontier	Romiti (2020)
8	Cross-chain	HTLC hash linking, timing/amount correlation, DP composition	Dwork (2006), P2C2T (2024)

Evidence fused via Dempster-Shafer (1967). Adversary-weighted scoring.

Adversary models

The same address has different scores depending on who's watching.

Model	Capabilities
casual	public block explorer
exchange	KYC + Chainalysis + ML classifiers
law-enforcement	subpoena + ISP data + full chain analysis
nation-state	backbone surveillance + Sybil + unlimited compute

etx analyse <addr> --adversary nation-state

Adversarial classification

Surrogate model approximating what chain analysis firms detect. Shows which features drive the classification and how to change it.

etx classify <txid>

Inverse-OSPEAD (novel)

Instead of attacking Monero rings (OSPEAD), construct the optimal ring that maximises adversary uncertainty. Selects decoys at indistinguishability ages where P_spend/P_decoy = 1.

etx ring <monero-hash> --optimise

Stack

packages/core/
  entropy/       Shannon, Dempster-Shafer, timing, amount, network, adversary models
  graph/         UTXO clustering (BFS + Tarjan + spectral Laplacian)
  fingerprint/   8-heuristic classifier, 10 recommendation types
  coinjoin/      Boltzmann entropy, subset sum, post-mix degradation
  monero/        ring analysis, inverse-OSPEAD, daemon RPC
  lightning/     routing privacy, Pareto frontier, mempool.space API
  crosschain/    bridge linking, DP composition, Wormhole/LayerZero API
  adversarial/   surrogate classifier, feature importance, evasion
  batch/         full wallet history, per-tx privacy timeline, degradation projection
  bitcoin/       Blockstream API (paginated, cached, rate-limited)

apps/cli/        8 commands, input validation, JSON piping, CSV export

Develop

npm install && npm run build
npm test           # 182 tests across 15 files
npm run cli -- analyse <address>

Contributing

See CONTRIBUTING.md.

Origin

Maths dissertation on Bitcoin cryptography. The dissertation asks "how does the cryptography work?" This tool asks "where does privacy actually break?"

License

MIT