SILS Framework: Strategic Liquidity Stability

• The SILS framework is a liquidity risk methodology that quantifies the impact of LP removal on market stability using counterfactual simulations with ETWL, anomaly detection, and LSIS.
• It replaces static, volume-based assessments with a dynamic model that evaluates how LP absence degrades market stability, reducing false positives and false negatives.
• It employs unsupervised anomaly detection to identify strategic LPs and supports real-time applications such as trader signals, risk dashboards, and automated liquidity interventions.

The SILS framework—Strategic Influence on Liquidity Stability—is a methodology for liquidity-provider risk analysis in concentrated-liquidity decentralized exchanges (CLMMs). It is defined as a shift away from static, volume-based analyses of liquidity-provider (LP) risk toward a dynamic, impact-focused model that treats LPs as strategic, systemic agents and asks: “How much market stability degrades if this LP were hypothetically removed?” The framework combines Exponential Time-Weighted Liquidity (ETWL), unsupervised anomaly detection, and the Liquidity Stability Impact Score (LSIS) to quantify each LP’s functional importance, with the stated aim of reducing false positives and uncovering false negatives that arise under nominal-size or raw-activity heuristics (RajabiNekoo et al., 25 Jul 2025).

1. Motivation and analytical framing

Traditional DeFi risk assessments, as characterized in the framework, label “whales” primarily by nominal size, such as an LP’s percentage of total value locked, or by raw activity, such as high mint/burn frequency. SILS identifies two failure modes in those approaches. The first is false positives: large or active LPs who, if removed, would barely change market stability. The second is false negatives: dormant LPs whose deep, well-positioned liquidity underpins price resilience (RajabiNekoo et al., 25 Jul 2025).

The framework therefore replaces static descriptors with a functional criterion based on market stability under counterfactual removal. This reorientation is presented as a move from classifying LPs as passive capital holders to characterizing them as dynamic systemic agents whose actions directly impact pool resilience. A plausible implication is that SILS is designed not merely for descriptive analytics, but for causal-style risk attribution at the LP level.

The market-stability quantity used in the framework is operationalized through price impact under a specified liquidity state. In the core formulation, the benchmark stability metric is the average price impact of a fixed set of synthetic swaps under the full liquidity state, and LP importance is measured by the degradation in that metric when a given LP’s events are excluded (RajabiNekoo et al., 25 Jul 2025).

2. Core quantities: ETWL and LSIS

SILS begins with a temporal representation of LP activity through Exponential Time-Weighted Liquidity. At time $t$, LP $i$’s ETWL is defined by

$$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$

Here, $t_k$ is the block timestamp or index of the $k$-th liquidity event, specifically a Mint or Burn, by LP $i$; $L_i(t_k)$ is the net change in active liquidity at event $k$; $\lambda$ is a positive decay rate; and the exponential factor gives recent contributions greater weight (RajabiNekoo et al., 25 Jul 2025).

This construction encodes two features simultaneously: the magnitude of liquidity contribution and its temporal recency. Because older events are down-weighted, ETWL is not a cumulative balance metric in the usual sense. Instead, it is a decayed activity-weighted liquidity profile. This suggests that the framework is intended to distinguish between currently consequential liquidity placement and historically large but presently less informative activity.

The principal causal score in SILS is the Liquidity Stability Impact Score. Let $V_{\text{baseline}}$ denote the chosen market-stability metric under the full liquidity state, and let $i$0 denote the same metric when LP $i$1’s events are excluded. SILS defines

$i$2

A positive $i$3 indicates that LP $i$4’s absence raises price impact and therefore degrades stability. Larger values indicate greater importance to pool resilience (RajabiNekoo et al., 25 Jul 2025).

The role of LSIS is conceptually distinct from ETWL. ETWL is a profile feature used for characterization and candidate selection, whereas LSIS is the framework’s counterfactual measure of systemic importance. The paper’s formulation makes LSIS the quantity that determines whether an LP is functionally critical, rather than merely large or active.

3. Strategic LP detection through unsupervised anomaly analysis

Before counterfactual scoring, SILS profiles every LP using two features: the LP’s ETWL score and the LP’s average price impact (PI) across historical trades or synthetic-swap approximations. For each LP $i$5, the feature vector is

$i$6

These vectors are then processed by an outlier-detection pipeline (RajabiNekoo et al., 25 Jul 2025).

The pipeline has four stated steps. First, SILS constructs the feature matrix from ETWL and average price impact. Second, it applies clustering or density estimation, specifically a Gaussian Mixture Model (GMM) or a robust clustering method such as DBSCAN. Third, it computes an outlier score, for example the LP’s Mahalanobis distance or an isolation-forest score, relative to the densest cluster. Fourth, LPs whose outlier score exceeds a chosen threshold, such as the 95th percentile, are marked as “strategic” or potential whales (RajabiNekoo et al., 25 Jul 2025).

This detection stage is not the final ranking mechanism. Rather, it acts as a screening layer that identifies candidates whose liquidity footprint and price-impact profile differ materially from the dense bulk of LP behavior. In the framework’s own workflow, the set of LPs tested with LSIS may be selected from top-$i$7 LPs by ETWL or anomaly score. This arrangement makes anomaly detection a practical prioritization mechanism within a larger causal assessment pipeline.

A plausible implication is that SILS separates two problems often conflated in prior heuristics: detecting atypical LP behavior and estimating actual systemic dependence on that behavior. In SILS, anomaly status alone does not establish criticality; the decisive quantity remains LSIS.

4. Counterfactual simulation and computational structure

Once a set of top-$i$8 LPs is selected, SILS computes LSIS through counterfactual removal. The framework first precomputes the baseline liquidity profile from the full event set $i$9, then evaluates the average price impact of a synthetic swap set $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$0 under that full profile. For each selected LP, the framework removes that LP’s events, rebuilds the liquidity profile, recomputes average price impact on the same synthetic swap set, and applies the LSIS formula (RajabiNekoo et al., 25 Jul 2025).

The paper specifies the computational complexity of this stage. Let $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$1 be the number of synthetic swaps and $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$2 the number of LPs tested. If liquidity-profile lookups use binary search over $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$3 price ticks, each average-price-impact run costs $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$4, yielding a total complexity of

$$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$5

The implementation guidance in the framework limits $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$6 to top ETWL candidates, for example $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$7, and $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$8 to $$ETWL_i(t) = \sum_{k: t_k \le t} L_i(t_k)\cdot \exp(-\lambda\cdot (t-t_k)).$$9, which is described as making the pipeline tractable. Reported optimizations include in-memory columnar data (Apache Arrow), pagination, and parallel execution (RajabiNekoo et al., 25 Jul 2025).

The counterfactual design is central to the framework’s interpretation. LSIS does not estimate what an LP has contributed historically in accounting terms; it estimates the incremental degradation in stability under removal from the reconstructed liquidity state. This makes the score a direct measure of dependence of the pool’s price-impact profile on a given LP’s events.

5. Empirical evaluation and comparison to baselines

The reported evaluation uses all Mint/Burn events for the USDC/WETH 0.05% Uniswap V3 pool on Ethereum Mainnet, covering blocks 12 376 729–21 001 766, or approximately 3.46 years (RajabiNekoo et al., 25 Jul 2025).

The framework is compared against three baselines:

Baseline	Definition
B1	Top 1% by static liquidity
B2	LP >1% of TVL
B3	High volume + high turnover ratio

The evaluation metrics are false positives (FP), defined as LPs flagged but with $t_k$0; false negatives (FN), defined as LPs missed but with $t_k$1; and precision/recall on “critical whale” labels (RajabiNekoo et al., 25 Jul 2025).

The reported results are specific. SILS reduced FP by ≈90% relative to B1/B2. It uncovered 15 “dormant but critical” whales with LSIS > 0.02 that B3 missed, corresponding to FN reduction ≈75%. Among the top 100 ETWL LPs, ranking by LSIS identified a small core (≈5 addresses) whose removal would raise price impact by thousands of percent; these are described as “linchpin whales” (RajabiNekoo et al., 25 Jul 2025).

These findings clarify the framework’s claimed distinction between capital size and functional importance. A large LP may not materially affect resilience, while a dormant LP with deep, well-positioned liquidity may be stability-critical. The evaluation therefore supports the framework’s stated objective of replacing binary or surface-level whale classifications with an impact-focused characterization.

6. Operational uses, limitations, and proposed extensions

The framework specifies three downstream application classes. The first is a protective oracle layer. In this design, a real-time oracle intercepts LP burn transactions, computes a quick stability check using an LSIS proxy and price-depth thresholds, and only permits withdrawals that do not breach pre-set stability budgets (RajabiNekoo et al., 25 Jul 2025).

The second is trader signals. By monitoring LSIS fluctuations, the oracle can flag upcoming periods of low liquidity resilience and alert market-making bots or human traders to widen spreads or hedge positions. The third is risk alerts and dashboards, where protocol risk managers subscribe to on-chain events tied to high-LSIS LPs, with rapid alerts triggering automated liquidity injections or parameter adjustments, including fees and cooldowns, to support pool health (RajabiNekoo et al., 25 Jul 2025).

The paper also states several limitations. SILS depends on complete, high-fidelity event logs (Mint/Burn). It incurs computational overhead for large $t_k$2 or high-resolution swap grids. It exhibits parameter sensitivity, particularly to the decay rate $t_k$3 and the anomaly threshold. Finally, it requires protocol-specific price-impact models for non-Uniswap V3 CLMMs (RajabiNekoo et al., 25 Jul 2025).

Its proposed future extensions are correspondingly practical. These include distributed GPU/cluster acceleration for real-time LSIS updates, automatic $t_k$4 tuning via Bayesian optimization or reinforcement learning, extension to cross-pool implications (multi-hop liquidity networks), and incorporation of adversarial game-theoretic simulations to predict LP strategic responses (RajabiNekoo et al., 25 Jul 2025).

Taken together, these elements position SILS as an LP-risk framework centered on counterfactual market-stability degradation rather than static exposure proxies. Its defining claim is not that large LPs are unimportant, but that functional importance must be established through impact on liquidity stability, with ETWL, anomaly detection, and LSIS serving distinct roles in that assessment (RajabiNekoo et al., 25 Jul 2025).