PaperlayerView exact revision on arXiv

Source-linked AI summary

Heterogeneous Cross-Chain Transaction Tracing for Solana Bridges via Candidate-Set Selective Decision

Wenjie Dou, Zheng Che, Meng Shen, Hanbiao Du, Qing Li, Yan Qiang

arXiv:2609.11413v1cs.CR

TL;DR

The paper tackles unreliable tracing of Solana-bound cross-chain transactions across heterogeneous ledgers, where missing standard event logs and uncertain target presence create semantic and association gaps. It formalizes four transaction modes and proposes SolTracer, which unifies execution semantics and selectively accepts or rejects candidates. SolTracer outperforms SOTA methods across three scenarios, improving open-world F1 by 20.16% over the strongest baseline.

  • Problem

    Existing tracing methods struggle to correlate Solana-bound transactions because Solana lacks standard event logs and cross-chain execution may not produce a valid one-to-one target correspondence.

  • Method

    SolTracer formalizes four Solana-bound transaction modes, maps disparate semantics into a unified event space, and uses candidate-set selective decision with multi-field consistency verification.

  • Results

    SolTracer outperforms SOTA methods across closed-world, open-world, and cross-source-chain generalization scenarios, improving open-world F1 by 20.16% over the strongest baseline.

  • Takeaways & Limitations

    The analysis characterizes Solana bridge dynamics through count-value divergence, cross-asset settlement shifts, target-observability gaps, and mechanism-dependent association reliability.

Abstract

from arXiv · show

Solana is a rapidly growing high-throughput blockchain platform that has attracted substantial liquidity and user activity. However, this expansion has also drawn the attention of illicit actors, who frequently leverage cross-chain bridges to route illicit funds onto Solana to obfuscate transaction lineage. Unlike EVM-compatible platforms, Solana features distinct execution dynamics and lacks standard event logs, creating severe semantic gaps that prevent existing tracing methods from reliably correlating cross-ledger transactions. In this paper, we formalize four types of Solana-bound cross-chain transaction modes and propose a candidate-set selective decision-based tracing method called SolTracer. SolTracer maps disparate execution semantics into a unified event space and employs candidate-set selective decision-making to reliably associate target transactions while abstaining when valid targets are absent. Extensive experiments demonstrate that SolTracer outperforms state-of-the-art (SOTA) methods across three representative scenarios: closed-world association, open-world association, and cross-source-chain generalization. In particular, under the challenging open-world setting with a 50% TA ratio, SolTracer improves the F1 score by 20.16% over the strongest SOTA baseline. Utilizing SolTracer, we conduct an empirical analysis on real-world cross-chain transfers to investigate ecosystem dynamics. Our analysis explores the stark count-value divergence across bridge mechanisms, the prevalence of cross-asset shifts, and the decoupling between on-chain settlement and explorer visibility.

I. INTRODUCTION

The paper addresses unreliable tracing of Solana-bound transfers across heterogeneous ledgers, where missing standard event logs and non-one-to-one settlement complicate correspondence. It formalizes four transaction modes, proposes SolTracer, and reports stronger association performance while analyzing bridge-ecosystem observability.

  • Motivation: Solana-bound tracing is difficult because heterogeneous execution models, absent standard event logs, and non-one-to-one settlement prevent direct cross-ledger field alignment.Existing approaches developed mainly for EVM-compatible chains can fail when transfers are delayed, batched, cross-asset, or unexecuted.
  • Method: SolTracer maps disparate transaction semantics into a unified event space and selectively associates candidates through multi-field consistency verification.Its candidate-set decision can abstain when the evidence does not support a valid target.
  • Contribution: The paper formalizes four Solana-bound transaction modes to characterize bridge funding workflows and execution dynamics.The abstraction supports analysis of how different bridge mechanisms expose transaction evidence.
  • Evaluation: SolTracer achieves the highest F1 scores among six SOTA methods in closed-world and open-world settings.The compared methods include AttrMatch, EntityMatch, MultiMatch, CONNECTOR, ABCTRACER, and ConneX.
  • Empirical analysis: An empirical analysis examines count-value divergence, cross-asset shifts, target observability gaps, and mechanism-dependent association reliability.The analysis uses associated real-world cross-chain transactions to study Solana bridge-ecosystem dynamics.

II. BACKGROUND AND RELATED WORK

Cross-chain tracing depends on source-chain intent, bridge verification, and target-chain settlement, but EVM and Solana expose these processes differently. Existing work leverages logs and metadata, whereas Solana’s fragmented records and absent candidate rejection create important limitations.

  • Cross-chain transactions: A cross-chain transfer involves source-chain intent, bridge authorization, and target-chain settlement across separate ledger components.The target chain may release escrowed assets, mint tokens, or trigger local contract calls after authorization validation.
  • EVM–Solana differences: EVM records bridge semantics in structured event logs, whereas Solana fragments evidence across programs, account roles, inner instructions, and balance deltas.Solana tracing therefore requires reconstructing evidence rather than relying on a single transaction receipt.
  • Related work: Existing tracing studies cover centralized and decentralized bridges using service APIs, transaction metadata, contract traces, event logs, and transfer values.The reviewed methods include heuristic deposit-withdrawal correlation and constraint-based association across ledger records.
  • Limitations: Existing methods assume homogeneous ledger evidence and rigid one-to-one mappings, so absent targets from delays, batching, or unexecuted transfers can yield false associations.Without a candidate-set reject option, ranking schemes still select a distractor when no valid target exists.

III. PROBLEM DEFINITION

The paper formulates Solana-bound tracing as selective association over a bounded candidate set that may not contain a valid target. It defines objectives for heterogeneous semantic matching and rejection, then organizes Solana transfers into four settlement modes.

  • Problem formulation: The tracing problem filters each source transaction into a bounded Solana candidate set, while allowing the ground-truth target to be absent because of delayed, batched, or unexecuted transfers.The target space includes an abstention outcome when no valid target is observed.
  • Selective decision: SolTracer selects the highest-scoring candidate only when a selective rule finds that its evidence dominates competing distractors; otherwise it outputs ⊥.This formulation explicitly supports target-present association and target-absent rejection.
  • Design goals: The design goals are to reconstruct unified heterogeneous semantics when the target is present and reject all candidates when reliable association is not guaranteed.The rejection objective is intended to prevent false-positive attributions.
  • Transaction modes: The paper formalizes Message Redemption, Pool Settlement, Solver Fulfillment, and Burn–Mint transactions according to their underlying funding topologies.Each mode is analyzed through its execution workflow and observability gaps.
  • Message Redemption Transactions: Message Redemption preserves source asset and amount semantics after verification, but Solana account transformations and indirect CPIs obscure recipient and transfer evidence.Recovering the target may require aggregating inner instructions and balance deltas beyond the top-level bridge invocation.

B. Pool-Settlement Transactions, PST

Pool Settlement and solver- or attestation-based modes weaken direct source-to-target continuity through amount changes, cross-asset fulfillment, separated funding, and split authorization evidence. These mechanisms create ambiguity for ledger-only association.

  • B. Pool-Settlement Transactions, PST: Pool Settlement draws Solana-side liquidity for recipient payouts, whose amounts can diverge from source values because of fees, slippage, liquidity, or conversions.The resulting amount relation is route-dependent rather than an exact source-target equality.
  • B. Pool-Settlement Transactions, PST: Shared pools can serve multiple transfers within one observation window, producing several payouts with similar timing and amount characteristics.This creates ambiguity among competing target candidates.
  • Solver Fulfillment Transactions: Solver Fulfillment records target parameters on the source chain while independent solvers fulfill them on Solana using proprietary liquidity.The target settlement is therefore decoupled from direct on-chain continuation of the source execution.
  • Solver Fulfillment Transactions: Cross-asset fulfillment makes raw asset equality unreliable, while solver-controlled funding removes direct asset-flow continuity between source and target.Intent or solver evidence may remain visible even when account-based links weaken.
  • Burn–Mint Transactions: Burn–Mint transfers split authorization and minting across off-chain attestation and on-chain execution phases, so ledger-only analysis may miss full provenance.Similar mint operations can also create candidate ambiguity when asset, amount, and timing overlap.

V. THE PROPOSED SOLTRACER

SolTracer traces Solana-bound cross-chain transactions through protocol identification, semantic reconstruction, and candidate-selective association. Its design combines protocol-aware evidence retrieval with selective abstention for unsupported or unreliable matches.

  • V. THE PROPOSED SOLTRACER: SolTracer comprises Bridge Protocol Identification, Semantic Event Reconstruction, and Candidate Selective Association modules.The overview presents these as the method's three main components.
  • V. THE PROPOSED SOLTRACER: Bridge Protocol Identification decodes source intent, retrieves compatible Solana fragments, and abstains when the transfer is unsupported.It uses protocol specifications and a thresholded compatibility score before downstream processing.
  • V. THE PROPOSED SOLTRACER: Candidate Selective Association scores bounded candidate sets using temporal, asset, amount, and participant consistency, then rejects unreliable sets.The reject option supports valid association when evidence is sufficient and abstention when candidates are unsupported or ambiguous.
  • V. THE PROPOSED SOLTRACER: Protocol specifications encode stable routing, asset, timing, amount, and participant-role invariants without caching ground-truth targets.These specifications form an extensible knowledge layer for protocol compatibility checks.

C. Semantic Event Reconstruction

Semantic Event Reconstruction converts heterogeneous EVM and Solana execution evidence into a common, provenance-aware event representation. Candidate filtering and selective scoring then use these events to support mode-specific matching without forced raw-field equality.

  • C. Semantic Event Reconstruction: Semantic Event Reconstruction integrates execution traces and state transitions into unified events containing route, participants, asset, amount, and provenance fields.Execution evidence includes receipts, logs, inner instruction trees, program logs, and balance transitions.
  • C. Semantic Event Reconstruction: Mode-specific adapters handle MRT, PST, SFT, and BMT through CPI/ATA extraction, pool deltas, solver payouts, and mint proofs.Each adapter records provenance for the evidence used to construct the event.
  • C. Semantic Event Reconstruction: Protocol-compatible filtering builds candidate sets using route, time, mapped asset, mode-specific amount, and participant-role checks.The matching design supports fee-adjusted PST payouts and cross-asset SFT fulfillment.
  • C. Semantic Event Reconstruction: SolTracer supplements pairwise ranking with candidate-set-level rejection based on score separation and semantic consistency.The rejector distinguishes empty candidate-space abstention from nonempty sets that fail the acceptance threshold.

VI. PERFORMANCE EVALUATION

The evaluation uses verified real-world EVM-to-Solana transfers spanning four source chains, four bridge mechanisms, and all four transaction modes. The resulting dataset supports systematic assessment of SolTracer across representative bridge settings.

  • VI. PERFORMANCE EVALUATION: 7,082 verified source–target pairs were retained from 38,317 retrieved bridge records across four EVM source chains and four representative bridges.The bridges cover Message Redemption, Pool Settlement, Solver Fulfillment, and Burn-Mint transaction modes.
  • VI. PERFORMANCE EVALUATION: Wormhole contributed 1,625 valid Message Redemption pairs after cross-file deduplication and dual-chain RPC verification.The pairs were collected across Ethereum, Arbitrum, Base, and Polygon PoS.
  • VI. PERFORMANCE EVALUATION: Allbridge contributed 1,341 valid Pool Settlement pairs after RPC verification, decoding, deduplication, and unstable-target filtering.The dataset includes source records from all four listed EVM chains.
  • VI. PERFORMANCE EVALUATION: deBridge contributed 1,597 valid Solver Fulfillment pairs after RPC verification of source orders and Solana fulfillment transactions.The pairs include records from Ethereum, Arbitrum, Base, and Polygon PoS.
  • VI. PERFORMANCE EVALUATION: Circle contributed 2,519 valid burn–mint pairs after filtering, deduplication, and semantic verification against Solana receive-and-mint transactions.The retained pairs span Ethereum, Arbitrum, Base, and Polygon PoS.

B. Experimental Settings

The experiments compare SolTracer with six matching baselines using precision, recall, F1, and group-level selective-rejection metrics. Validation varies target presence and absence across candidate groups while separating training, validation, and test data.

  • B. Experimental Settings: The comparison includes AttrMatch, EntityMatch, MultiMatch, CONNECTOR, ABCTRACER, and ConneX after adaptation to unified semantic-event fields.The baselines span heuristic, constraint-based, retrieval, and learning-based approaches.
  • B. Experimental Settings: Performance is measured with precision, recall, and F1, while group-level metrics quantify abstention when valid targets are absent.The group metrics are Pg, Rg, and F1g.
  • B. Experimental Settings: Candidate groups are labeled Target-Present or Target-Absent, with controlled masking used to evaluate open-world rejection under different class ratios.Closed- and open-world experiments use Ethereum-to-Solana data, while generalization changes the source-chain setting.
  • B. Experimental Settings: The candidate ranker uses a random forest over 14 semantic features, while the rejector excludes candidate count and minimum time distance to avoid structural leakage.Main experiments are repeated with five fixed random seeds.

C. Sensitivity Analysis

SolTracer is evaluated across closed-world, open-world, and cross-source-chain settings, with selective thresholds calibrated for pair acceptance and candidate-set rejection. It achieves the strongest reported association and rejection performance, including a 20.16-point advantage at 50% target absence.

  • Sensitivity Analysis: τp = 0.94 maximizes candidate-pair F1, while τr = 0.227 maximizes group-level rejection F1.The rejection threshold achieves Pg = 89.74%, Rg = 87.50%, and F1g = 88.61%.
  • Sensitivity Analysis: Pair-level filtering and set-level abstention govern distinct error modes and remain stable when calibrated independently.The two thresholds separately control individual association acceptance and whether a candidate pool should return any association.
  • Closed-World Association: 98.13% overall F1 lets SolTracer outperform ConneX by 12.30 percentage points in closed-world association.It also reports 99.46% precision and 96.84% recall.
  • Open-World Association: At 50% target absence, SolTracer’s candidate-level F1 advantage over the strongest baseline reaches 20.16 percentage points.Its candidate F1 decreases from 96.50% at 10% target absence to 93.87% at 50%, while rejection F1g rises from 94.87% to 97.19%.

F. Evaluation on Generalization

SolTracer is evaluated through module ablations and cross-source-chain transfer from Ethereum to Arbitrum, Base, and Polygon PoS. It retains strong performance across unseen source chains, while each major component contributes to association or rejection quality.

  • Cross-Source-Chain Generalization: 97.79% candidate F1 and 91.87% rejection F1g give SolTracer the strongest overall cross-source-chain generalization.It exceeds the strongest baselines by 11.33 and 10.32 percentage points, respectively.
  • Cross-Source-Chain Generalization: SolTracer’s candidate F1 reaches 96.06%, 98.79%, and 98.53% on Arbitrum, Base, and Polygon PoS, respectively.The corresponding gains over the strongest baseline are 12.27, 10.72, and 5.11 percentage points.
  • Ablation Study: Without protocol-guided filtering, candidate-matching F1 drops by 20.93%, while rejection F1g falls from 94.87% to 37.86%.Generic raw-attribute matching fails under cross-chain conversions and filters out supported targets.
  • Ablation Study: Replacing mode-guided event reconstruction reduces candidate F1 to 96.37% and set-level rejection F1g by 2.63%.Mode-specific account roles and inner invocations provide semantic cues for candidate distinction and support assessment.
  • Ablation Study: Logistic Regression reaches only 77.98% candidate F1 and 62.92% rejection F1g, whereas XGBoost preserves candidate F1 but loses 2.98% in rejection F1g.The results favor nonlinear modeling and identify Random Forest as the most balanced classifier.
  • Ablation Study: Removing candidate-set rejection lowers candidate F1 from 96.50% to 92.05% and eliminates identification of target-absent groups.The ablation supports selective abstention as a safeguard against forced false-positive attributions.

VII. EMPIRICAL INSIGHTS ON SOLANA CROSS-CHAIN TRANSACTIONS

SolTracer’s real-world analysis reveals that transaction counts poorly reflect transferred value, source and settlement assets often differ, and explorer records miss settled Solana targets. Selective tracing separates settled-but-unlinked, ambiguous, and unresolved transfers instead of forcing unsupported links.

  • Count–Value Divergence: SFT represents 65.86% of transactions but 15.53% of volume, whereas BMT represents 16.28% of transactions but 51.03% of volume.MRT also has a higher volume share than transaction share, while PST is minimal on both measures.
  • Cross-Asset Settlement Patterns: Non-USDC assets rise from 9.10% of source-chain deposits to 52.76% of Solana settlements.USDC falls from 90.90% to 47.24%, while native SOL comprises 36.13% of settlement assets.
  • Target Observability: Among 47,117 indexed records, 20,202 transfers, or 42.88%, contain only an origin deposit without an explicit target-side link.The analysis classifies these source-only records by replaying them against bounded Solana candidate pools and mode-specific evidence.
  • Target Observability: SolTracer recovers 2,201 targets from source-only records, including 2,199 settled before the collection cutoff.The recovered targets comprise 1,493 Circle CCTP and 706 Wormhole WTT transactions, indicating that most missing links are indexing omissions.
  • Target Observability: 184 Circle CCTP transfers remain ambiguous because multiple concurrent mint operations satisfy identical temporal, recipient, and amount constraints.The finding supports set-level selective rejection when observable evidence cannot identify a unique target.
  • Target Observability: The remaining 14,848 Circle CCTP and 2,969 Wormhole WTT records have no valid match within the observation window, so SolTracer abstains.These records may reflect unsupported routes, long-tail finality delays, or genuinely unexecuted transactions.

D. Mode-Dependent Association Reliability

Association and rejection reliability varies systematically across the four Solana bridge modes because each preserves different discriminative evidence. These profiles motivate mode-specific feature modeling and calibrated abstention for open-world tracing.

  • Fig. 10 summarizes mechanism-specific candidate-set evidence retention across time, amount, asset, recipient, and role cues under target-present and target-absent settings.
  • MRT preserves recipient and redemption-role cues, while PST introduces slippage and fee-related amount deviations alongside overlapping shared-pool payouts.
  • SFT has the weakest direct linkage because solver fulfillment decouples funding accounts and permits cross-asset conversions.
  • Association and rejection reliability depends heavily on the bridge mechanism, requiring mode-specific feature modeling and calibrated abstention.
  • The analysis covers four Solana-bound transaction modes—MRT, PST, SFT, and BMT—defined by their settlement mechanisms.
  • SolTracer outperforms state-of-the-art methods across closed-world, open-world, and cross-source-chain generalization scenarios.
Loading 2609.11413v1…