Skip to content
Crypto.world

Research

The Anonymity Gap: Understanding Real Privacy in Shielded UTXO-based Protocols for DeFi

Mixer-style privacy collapses anonymity to a one-hop matching problem between deposits and withdrawals, but shielded UTXO protocols let funds keep moving, splitting, merging, and being re-spent…

The Anonymity Gap: Understanding Real Privacy in Shielded UTXO-based Protocols for DeFi
arXiv
2608.22987
Published
2026-08-24
Authors
Hanze Guo, Stefanos Chaliasos, Yebo Feng, Jiahua Xu

AI summary

Overview

Research area: Blockchain privacy — specifically, on-chain anonymity analysis of shielded UTXO-based privacy protocols used for DeFi (Railgun, Hinkal), set within the broader cryptography-and-security literature on mixer tracing and address-linkage analysis.

Technical level: Advanced. The paper builds a formal three-layer system model (note, proof, transaction) with set-theoretic definitions, pseudocode pruning strategies, a fixpoint propagation procedure, and a recursive closure operator. Readers need familiarity with commitments, nullifiers, Merkle trees, zero-knowledge proofs, and anonymity sets.

Scope (one sentence): The paper measures, using only public on-chain traces and no heuristics, how much the set of possible originating ("shielding") addresses shrinks for transactions that release funds from Railgun and Hinkal shielded pools, across the complete histories of all four Railgun production deployments and five Hinkal pools on six EVM chains.

What This Paper Is About

Mixer-style privacy collapses anonymity to a one-hop matching problem between deposits and withdrawals, but shielded UTXO protocols let funds keep moving, splitting, merging, and being re-spent inside a hidden state before they ever exit. That means the real question is not "how big is the pool" but "which shielding addresses could this withdrawal have come from, once every publicly visible constraint is followed backward through the hidden state?"

The authors build a framework to answer that question rigorously and non-heuristically, then apply it to 186,356 unshielding spend transactions to quantify how much anonymity deployed protocols actually deliver versus the naive pool-size baseline.

Key Contributions

  1. A layered system model with formal privacy metrics. The paper formalizes shielded UTXO protocols across a note (UTXO) layer, a proof layer, and a transaction layer, and defines two metrics: proof-level Commitment Set Size (how many historical commitments remain feasible as hidden inputs to a given proof) and transaction-level Anonymity Set Size (how many shielding addresses could have originated the funds released by one unshielding transaction), each with a temporal baseline.

  2. A non-heuristic recursive provenance pipeline. Four cumulative pruning stages — tree number, proof-root coverage, token consistency, and value feasibility — are applied to each proof's candidate set, followed by cross-proof global propagation to a fixpoint, and then a closure-based backward lift that traces surviving commitments to the shield transactions and addresses that created them.

  3. A full-population empirical measurement. The analysis covers 313,492 proofs and 259,049 spend transactions (186,356 of them unshielding) across Railgun and Hinkal, reporting mean Anonymity Set Size reductions of 40.1%–59.0% for Railgun chains and 48.9%–57.4% for Hinkal chains relative to temporal baselines.

  4. An eight-pattern anonymity-loss taxonomy spanning proof-level specialization, proof-structure effects, and transaction-level aggregation, supported by case studies and distilled into implications for user behavior, privacy-protocol interfaces, and protocol design.

Main Findings

  • Anonymity falls well short of pool size. Mean Anonymity Set Size shrinks by 40.1%–59.0% relative to each Railgun deployment's temporal baseline, and by 48.9%–57.4% across the Hinkal chains, using only public protocol traces and constraints.

  • A small but real population of near-identified transactions. 3,679 transactions retain at most 10 candidate addresses, including 1,228 singletons. Broken down: 2,968 Railgun transactions and 711 Hinkal transactions at most ten addresses, of which 920 and 308 respectively are singletons.

  • Token constraints dominate. Public token constraints are described as the strongest and most stable source of pruning in both protocols.

  • Other constraints are protocol- and history-dependent. The effects of tree number, proof roots, and value constraints vary with protocol design and historical state rather than being uniformly strong.

  • Privacy demand migrated rather than disappeared. After the August 2022 OFAC sanctions on Tornado Cash, daily transactions on Tornado Cash Classic and Tornado Cash Nova fell by 97% and 91% respectively, while Railgun's share of transactions among major Ethereum privacy protocols rose from 13% in 2022 to 71% in 2025.

  • Railgun's scale at the cutoff. Railgun's TVL was approximately $80.23 million (from DefiLlama, July 30, 2026); the analysis covers all four of its production deployments — Ethereum, Arbitrum, Polygon, and BNB Chain — from initial deployment on each chain.

  • Hinkal's deployment landscape is fragmented. Among 39 independently deployed pool states across ten chains, 37 are nonempty and 23 of those contain fewer than 1,000 commitments; because each deployment keeps an independent commitment tree, these pools cannot be concatenated into a single anonymity domain.

  • Recursive provenance differs qualitatively from mixer tracing. Because a single spend transaction can aggregate multiple proofs, and hidden value can be re-spent repeatedly before any public exit, the feasible source set propagates across layers rather than being a one-hop deposit-withdrawal match.

Methodology in Plain English

The authors treat each unshielding transaction as a question: which addresses that ever put money into this pool could this money have come from? They then shrink the answer using only things anyone can see on-chain.

They start with a temporal baseline — every commitment that already existed when the transaction began executing. Anything created inside the same transaction is excluded, since a transaction cannot spend its own fresh outputs.

Next they apply four filters, cumulatively, to each proof:

  1. Tree number — a private spend proves membership in one specific public Merkle tree, so commitments in other trees are impossible.
  2. Proof root — a spend is verified against a specific public root snapshot, so commitments inserted after that root's largest covered leaf index are too new to be included.
  3. Token — the pool holds multiple assets, and token information leaks through creation traces and revealed outputs. Candidate commitments whose possible token interpretations cannot supply the proof's required number of inputs of a permitted token get dropped. When token branches are publicly indistinguishable, the method keeps all of them rather than guessing.
  4. Value — arithmetic constraints (exact revealed amounts, or bounded intervals when only part of the value is public) eliminate candidates whose values cannot participate in any feasible balance. This step is deliberately conservative: it removes a candidate only when public constraints rule it out.

Then comes cross-proof propagation: proofs constrain each other, because a mandatory input consumed by one proof is unavailable to another, and a newly created commitment may carry a narrower token or value interpretation forward. The procedure revisits only affected proofs until nothing more can shrink.

Finally, closure-based lifting walks backward from each surviving commitment through the history of hidden-state transitions — following creating proofs and creation traces — collecting all terminal shield transactions that could have introduced the funds. Mapping those transactions to their public entry addresses gives the Anonymity Set Size for the transaction, and the union is taken across all proofs in it that released value.

Deliberately excluded: address clustering into users or entities, behavioral timing patterns, probabilistic weighting of candidates, and any off-chain attribution. This is what makes the analysis "non-heuristic" — it isolates anonymity loss caused by the protocol's own public constraints. The authors note that their reported sets therefore upper-bound the exact jointly feasible candidate sets, since the value filter does not enumerate every concrete input tuple.

Why This Matters

Impact on research. Prior applied work on deployed privacy relied heavily on heuristics — address clustering, timing, value-pattern matching, decoy selection. This paper establishes a non-heuristic baseline by asking only what public protocol traces mathematically rule out. That baseline is useful either way: heuristic or probabilistic layers can be added on top and their marginal contribution measured, rather than being entangled with what the protocol already leaks. It also shifts the analytical object from "pool size" to "provenance," which is the correct abstraction for protocols where value keeps moving inside the hidden state.

Real-world applications:

  • User risk assessment — a user of Railgun or Hinkal can understand that a withdrawal does not inherit the full pool's anonymity, and that certain transaction shapes leave very small candidate sets (in this data, sometimes a single address).
  • Wallet and interface design — wallets and relayers that construct shielded transactions can surface the effective anonymity set before a user unshields, rather than a misleading pool-size number.
  • Protocol design iteration — developers can see which public constraints (especially token encoding and revealed output values) do the most pruning, and decide what to reveal or hide at the design level.
  • Compliance and audit reasoning — the taxonomy of anonymity-loss patterns gives a structured vocabulary for describing how much ambiguity a given transaction actually retains, useful for exchanges and auditors reasoning about provenance.

Industry relevance. Railgun is described as a leading DeFi-facing privacy protocol, with TVL of roughly $80.23 million at the cutoff and a share of transactions among major Ethereum privacy protocols rising from 13% in 2022 to 71% in 2025. Shielded UTXO protocols are described as becoming core privacy infrastructure for DeFi, so measurements of their actual — not nominal — anonymity bear directly on how much privacy users are really buying.

Future Directions

  • Extending to entity level. The metrics are address-level by construction. Lifting them to users requires address attribution, typically heuristic or off-chain; the authors explicitly position their work as the clean baseline such an extension would build on.

  • Adding probabilistic weighting. The paper counts feasible candidates without ranking them. Turning these sets into entropy-based anonymity measures requires an explicit posterior over candidates from behavioral priors, attribution models, or off-chain information — left open.

  • Improving the value-feasibility operator. The current filter is conservative and does not enumerate concrete multi-input tuples during fixpoint updates, so reported sets upper-bound the exact feasible sets. Exact joint solving could tighten results.

  • Design guidance for token and output revelation. Since token constraints proved the strongest and most stable pruning source in both protocols, a concrete open question is how protocols can encode multi-asset notes and reveal output values without handing observers such a powerful constraint — and how users should structure transfers to avoid collapsing into the small-set tail.

  • Whether other deployed protocols behave the same way. Hinkal's fragmented deployment landscape (39 pool states, 23 with fewer than 1,000 commitments) shows the results depend heavily on how a protocol shards its anonymity domain; how coordinated cross-chain generations should be structured is unresolved.

Target Audience

Researchers in blockchain privacy and applied cryptography; protocol designers and cryptographers working on shielded pools; security analysts and auditors who trace illicit flows and need to know what public constraint propagation can and cannot establish; and advanced DeFi users or wallet engineers who want a realistic rather than nominal picture of the anonymity a shielded pool provides.

Authors’ abstract

Shielded UTXO-based protocols are becoming a core form of privacy infrastructure for DeFi. Unlike mixers that organize privacy mainly around deposits and withdrawals, these protocols allow assets, once inside the shielded pool, to continue moving and being re-spent within the hidden state, and to become public only when users withdraw or interact with public DeFi protocols. Their anonymity is therefore no longer a flat pool-size problem, but a provenance problem that…

The full abstract on arXiv
Read the original paper