# q-fin.CP Published articles for q-fin.CP. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Towards Robust Reinforcement Learning Market Making: Lessons from Loss Function Experimentation DevFeed: [Towards Robust Reinforcement Learning Market Making: Lessons from Loss Function Experimentation]() Original publisher: [Read original article]() Author: Siddharth Nand, Bora Guney Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Reinforcement learning](), [Machine learning]() Tags: [algorithmic-trading](), [cs-lg](), [evaluation](), [experimentation](), [financial-markets](), [inventory](), [management](), [market-making](), [q-fin-cp](), [q-fin-tr](), [reinforcement-learning](), [risk](), [risk-management](), [robust]() ### AI overview The abstract reports that reinforcement learning market-making strategies with penalties for inventory and cash management were more resilient and adaptable than profit-maximizing approaches in the paper's empirical evaluation. It presents balancing profit and risk management as important, but provides no details on the loss formulations, experimental setup, or measured outcomes. ### Source excerpt arXiv:2611.09012v1 Announce Type: new Abstract: Market makers play a vital role in ensuring efficient trade within financial markets. Recent literature has delved into training market-making systems using reinforcement learning. This paper delves into the impact of various loss function formulations on risk management behavior within reinforcement-based market making. Through empirical evaluation, we contrast profit-maximizing strategies with risk-averse approaches, finding that penalties for inventory and cash management foster more resilient and adaptable market-making strategies. Our results underscore the significance of striking a balance between profit maximization and risk management in algorithmic trading, providing actionable insights for refining strategies in practical settings. ## FINCOSMOS: An Open-World Evaluation Paradigm for Agents in Finance DevFeed: [FINCOSMOS: An Open-World Evaluation Paradigm for Agents in Finance]() Original publisher: [Read original article]() Author: Yupeng Cao, Xueqing Peng, Lingfei Qian, Haohang Li, Yangyang Yu, Anke Xu, Wenbo Cao, Yan Wang, Yuechen Jiang, Yuyan Wang, Zhuohan Xie, Yankai Chen, Zehan Li, Jimin Huang, Xue Liu, Jian-Yun Nie, Sophia Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Finance](), [AI Chat]() Tags: [agents](), [backtest](), [benchmarks](), [coding-agents](), [cs-ai](), [cs-cl](), [data](), [evaluation](), [finance](), [open](), [persistent-memory](), [point-in-time-correctness](), [q-fin-cp](), [world]() ### AI overview FINCOSMOS evaluates financial agents in open-ended tasks where they must elicit a user's hidden strategy, obtain Web data, respect a data cutoff, and deliver verifiable backtests. Its grading checks point-in-time correctness, recomputes reported metrics, compares the delivered strategy with the hidden specification, and requires user satisfaction plus passing 15 veto checks; it does not need a reference answer. In 150 sessions across four coding agents and one model without tools, none was accepted. Every user-approved session failed a program check, and agents elicited only 54-64% of unstated details. Human experts attributed 88.33% of failures to agents. With FINCOSMOS feedback, an agent with persistent memory learned all ten tasks. ### Source excerpt arXiv:2611.07464v1 Announce Type: new Abstract: Language agents increasingly work for users on the open Web, and finance is where trusting them is hardest. The data must come from the Web, the strategy lives in the user's head, and correctness depends on time, so an agent can fail silently: a strategy the user never asked for, unverified data, or a backtest that uses the future yields a report as convincing as a correct one. Existing benchmarks close the world somewhere, fixing the task, the data, or the timeline, and thereby remove these failures instead of measuring them. We introduce FINCOSMOS, an open-world evaluation paradigm for agents doing financial work. A simulated user holds a hidden strategy sampled from a compositional grammar and reveals it only when asked. The agent receives no data, so it must acquire data from the Web, build its own backtest under a data cutoff, and deliver verifiable results. Grading needs no reference answer. Programs verify point-in-time correctness and recompute every reported metric, the delivered strategy is checked against the hidden specification, and a run is accepted only when the user is satisfied and none of 15 veto checks fails. Across four coding agents and a model without tools in 150 sessions, no session is accepted: every session that the simulated user signed off fails a program check, and agents raise only 54--64\% of the details they were not told. Human experts attribute 88.33\% of failures to the agents. With FINCOSMOS feedback, an agent with persistent memory learned all ten tasks. ## Deep Neural Operator Learning for Probabilistic Models DevFeed: [Deep Neural Operator Learning for Probabilistic Models]() Original publisher: [Read original article]() Author: Erhan Bayraktar, Qi Feng, Zecheng Zhang, Zhaoyu Zhang Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Deep learning](), [Neural Network](), [Artificial Intelligence]() Tags: [approximation](), [architecture](), [cs-lg](), [deep](), [general](), [global](), [learning](), [model](), [network](), [neural](), [operator](), [operator-learning](), [option](), [options](), [pricing](), [probabilistic](), [problems](), [processes](), [q-fin-cp](), [sde](), [space]() ### AI overview The paper proposes a deep neural-operator framework for probabilistic models and proves universal approximation with explicit network-size bounds under global Lipschitz conditions. Its stochastic-process assumptions are integrability and general tail-probability conditions. The authors verify these conditions for European and American option pricing in an FBSDE framework, which covers operators arising from parabolic PDEs, including problems with free boundaries. A numerical example for a basket of American options shows the model producing optimal stopping boundaries for new strike prices without retraining. ### Source excerpt arXiv:2511.07235v2 Announce Type: replace Abstract: We propose a deep neural-operator framework for a general class of probabilistic models. Under global Lipschitz conditions on the operator over the entire Euclidean space-and for a broad class of probabilistic models-we establish a universal approximation theorem with explicit network-size bounds for the proposed architecture. The underlying stochastic processes are required only to satisfy integrability and general tail-probability conditions. We verify these assumptions for both European and American option-pricing problems within the forward-backward SDE (FBSDE) framework, which in turn covers a broad class of operators arising from parabolic PDEs, with or without free boundaries. Finally, we present a numerical example for a basket of American options, demonstrating that the learned model produces optimal stopping boundaries for new strike prices without retraining. ## FinMultiHopQA: A KG-Grounded Multi-Hop Financial QA Benchmark for Evidence-Driven Reasoning DevFeed: [FinMultiHopQA: A KG-Grounded Multi-Hop Financial QA Benchmark for Evidence-Driven Reasoning]() Original publisher: [Read original article]() Author: Abhinav Arun, Reetu Raj Harsh, Bhaskarjit Sarmah, Stefano Pasquali Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [llm-reasoning](), [Finance](), [база знаний]() Tags: [benchmark](), [cs-ai](), [financial](), [knowledge-graph](), [q-fin-cp](), [qa](), [reasoning](), [reproducibility]() ### AI overview FinMultiHopQA provides 15,600 qualitative and quantitative financial QA questions grounded in FinReflectKG, a temporal knowledge graph built from SEC 10-K filings with more than 17.5 million triples. Its questions instantiate reproducible two- and three-hop motifs across filing sections, fiscal years, and companies, with linked provenance. In a controlled comparison, the benchmark keeps evidence constant while varying its presentation across six formats. The reported results show lower judged correctness with less-structured presentations, especially for quantitative cross-company questions and three-hop branch motifs; KG-linked presentation also uses fewer input tokens by design. The release includes provenance annotations, model outputs, evaluation records, and technical documentation. ### Source excerpt arXiv:2510.02906v2 Announce Type: replace-cross Abstract: Multi-hop financial question answering often requires composing evidence distributed across filing sections, fiscal years, and peer companies. Prior financial QA benchmarks address parts of this problem individually, but none jointly support controlled multi-hop taxonomies, document topologies, motif-level reproducibility, and provenance-linked auditability at scale. We introduce FinMultiHopQA, a benchmark of 15,600 qualitative and quantitative questions, grounded in FinReflectKG, a temporally indexed financial knowledge graph derived from SEC 10-K filings with 17.5M+ triples. The pipeline instantiates statistically salient 2-3 hop typed KG motifs across intra-document, inter-year, and cross-company regimes. Beyond closing this gap, the same construction lets us hold each question's underlying evidence constant while varying only its presentation - embedding it in noisy surrounding context or changing its order - giving a controlled instrument to study a secondary question at scale: using six such presentations, we test whether reasoning LLMs' performance depends on evidence structure. Results confirm this: despite their reasoning capacity, judged correctness remains consistently lower under less-structured presentations of the same evidence - most sharply for quantitative cross-company reasoning and 3-hop branch motifs - indicating that reasoning alone does not fully offset how evidence is structured; KG-linked presentation also uses fewer input tokens by design. We release the benchmark, provenance annotations, model outputs, evaluation records, and supporting technical documentation. ## The Digital-Energy Recomposition of the Kitchin Cycle: Exploratory Evidence from U.S. IT and Irish Data-Centre Series (1995-2025) DevFeed: [The Digital-Energy Recomposition of the Kitchin Cycle: Exploratory Evidence from U.S. IT and Irish Data-Centre Series (1995-2025)]() Original publisher: [Read original article]() Author: Bartosz Ziollko, Adam Niedzialkowski, Bartosz Szyma Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Sustainable Computing]() Tags: [2025](), [adjustment](), [aggregate](), [ai](), [assets](), [boundary](), [capacity](), [cloud](), [compute](), [correlation](), [cs-ai](), [data](), [data-centre](), [decoupling](), [digital](), [econ-gn](), [electricity](), [energy](), [hyperscaler](), [it](), [physics-soc-ph](), [q-fin-cp](), [q-fin-ec](), [q-fin-gn](), [recomposition](), [series]() ### AI overview The analysis reports that the U.S. IT-GDP correlation has a statistically significant break at 2014Q3, falling from +0.45 to −0.12; the break predates the imposed Cloud-AI boundary, which is not significant after adjustment. In Ireland, data-centre electricity correlates with Real GDP during the AI era (r=+0.41, N=16), but the relationship falls to about zero using Real GNP, consistent with shared multinational tax effects. Cross-correlations also indicate GDP tends to precede data-centre load, making compute capacity coincident or moderately lagging rather than a leading indicator. The authors describe the findings as exploratory given small AI-era samples and the lack of quarterly GNI* data. ### Source excerpt arXiv:2610.33122v1 Announce Type: cross Abstract: Using quarterly FRED and CSO Ireland data over 1995Q1-2025Q4, we examine whether U.S.\ IT investment and Irish data-centre electricity are recomposing the Kitchin cycle around energy- and compute-backed assets. Three findings stand out. First, a Quandt-Andrews sup-Wald test locates a significant break in the U.S.\ IT-GDP correlation at 2014Q3 (sup-W $= 9.69$; $r$ falls from $+0.45$ to $-0.12$); the Fisher $r$-to-$z$ test survives Bonferroni adjustment ($z = +3.11$, $p_{\text{Bonf}} = 0.006$), whereas the analogous test on the imposed Cloud-vs-AI boundary does not ($p_{\text{Bonf}} = 0.164$), so ``Strategic Decoupling'' pre-dates the AI Era by approximately six years. Second, in Ireland the synchronous AI-Era correlation between metered data-centre electricity and Real GDP is large ($r = +0.41$, $N = 16$) but \emph{collapses to $r \approx 0$ when Real GDP is replaced by Real GNP}, suggesting the headline is largely an artefact of the shared multinational tax-structure footprint between hyperscaler operations and Irish gross output. Third, the Irish cross-correlation profile is asymmetric in the opposite direction to a leading-indicator reading: the largest positive coefficients cluster on the $k < 0$ side (GDP precedes data-centre load), so compute capacity is better read as coincident or moderately lagging. Given the small AI-Era samples and the absence of a quarterly multinational-adjusted aggregate such as GNI*, all coefficients are exploratory; subject to these caveats, the U.S.\ Kitchin cycle is best dated to 2014Q3, and the Irish Energy-Data Nexus is dominated by multinational firm activity. ## CosmosAlpha: Multi-Relational Graph Search with Pareto-Guided Exploration for LLM-Driven Alpha Mining DevFeed: [CosmosAlpha: Multi-Relational Graph Search with Pareto-Guided Exploration for LLM-Driven Alpha Mining]() Original publisher: [Read original article]() Author: Zhan-He Wang, Zi-Yang Lang, Yi-Yang Li, Qing-Yu Lan, Yun-Song Piao Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Data Mining & Modeling](), [Large Language Model](), [Graphs]() Tags: [abstract](), [account](), [alpha](), [arxiv](), [based](), [best](), [can](), [code](), [correlation](), [cross](), [cs-ai](), [cs-lg](), [csi](), [data](), [discovery](), [diversity](), [exploration](), [graph](), [llm](), [mining](), [q-fin-cp](), [search](), [with]() ### AI overview CosmosAlpha searches for predictive financial factors by linking them in a graph that captures generation lineage, expression similarity, and cross-sectional exposure correlation. Its Graph Flow mechanism propagates search utility across related factors, while Pareto-guided selection balances quality, fertility, and diversity. On CSI 300 and CSI 500, CosmosAlpha-Graph achieved the best mean rank across seven predictive and portfolio metrics and outperformed the controlled CosmosAlpha-MCTS counterpart on every reported metric. These results support a consistent advantage for relational graph search in the reported evaluation. ### Source excerpt arXiv:2611.04555v1 Announce Type: cross Abstract: Discovering predictive alpha factors from noisy financial data is a fundamental yet challenging problem in quantitative finance. LLM-driven formulaic alpha mining offers an attractive paradigm by combining interpretable symbolic expressions with iterative generation and refinement guided by quantitative feedback. However, existing generation-centric and tree-based search strategies primarily organize factors through generation lineage, leaving structural and statistical relationships across search branches underutilized. Consequently, information remains localized along parent--child trajectories, which can confine exploration to redundant regions. Moreover, effective factor discovery must account for multiple distinct objectives that cannot be adequately captured by a single search criterion. We introduce CosmosAlpha, a graph-based LLM factor-mining framework that represents discovered factors as a multi-relational graph encoding generation lineage, expression-structure similarity, and cross-sectional exposure correlation. A Graph Flow mechanism propagates search utilities across the graph, allowing each factor to be assessed using relational evidence from nearby factors rather than in isolation. Pareto-guided selection jointly considers quality, fertility and diversity, explicitly modeling their trade-offs within a multi-objective search framework. Across CSI 300 and CSI 500, CosmosAlpha-Graph achieves the best mean rank over seven predictive and portfolio metrics and outperforms its controlled CosmosAlpha-MCTS counterpart on every reported metric, showing a consistent advantage for relational graph search in LLM-driven factor mining. The source code is available at https://github.com/wzhmtldswz/CosmosAlpha ## Beyond Binary Detection: Type-Aware Hallucination Analysis and Hierarchical Detection in Financial Question Answering DevFeed: [Beyond Binary Detection: Type-Aware Hallucination Analysis and Hierarchical Detection in Financial Question Answering]() Original publisher: [Read original article]() Author: Xiongyu Chen, Sizhen Zhu, Zhuoer Wang Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Hallucination detection](), [AI Chat]() Tags: [analysis](), [benchmark](), [categories](), [classes](), [client](), [compliance](), [cs-ai](), [detection](), [diagnose](), [diagnosis](), [errors](), [evaluation](), [financial](), [generation](), [grounding](), [hallucination-detection](), [hallucinations](), [input](), [investment](), [q-fin-cp](), [question](), [type]() ### AI overview On PHANTOM, a benchmark built from SEC filings, evaluating hallucination detectors by error type revealed weaknesses hidden by binary accuracy. Across 15,443 paired faithful and hallucinated answers, hallucinated answers were more than twice as long, while lexical grounding separated classes only because faithful answers were bimodal. In 4,086 usable examples spanning nine financial hallucination categories, an LLM detector achieved 80.2% binary accuracy but 44.7% subtype accuracy, never assigned two categories, and missed 5.3%-46.8% of hallucinations depending on category. Misses were unrelated to input length and increased with the answer's divergence from the correct one. A proposed two-stage detector first identifies hallucinations, then assigns a subtype; it reached 95.6% binary and 56.1% subtype accuracy. The abstract reports generation success varied from 66.3% to 100% across categories. ### Source excerpt arXiv:2611.02258v1 Announce Type: new Abstract: Large language models are increasingly used to answer questions over financial filings, where an unfaithful answer can mislead investment, compliance, or client decisions. Hallucination detectors are a common safeguard, but they are usually evaluated with a single binary score that reveals neither which kinds of errors analysesed nor why. This paper studies hallucination detection on PHANTOM, a long-context benchmark built from SEC filings, at the level of hallucination type. We first analyze 15,443 paired hallucinated and faithful answers without any detector and find that hallucinated answers are more than twice as long as faithful ones, and that lexical grounding separates the two classes only because faithful answers are bimodal. We then generate hallucinations in nine financially motivated categories, confirm each with a separate verification pass, and diagnose an LLM-based detector on 4,086 usable examples. Generation success varies from 66.3% to 100% across categories. The detector reaches 80.2% binary accuracy but only 44.7% subtype accuracy, never assigns two of the nine categories, and misses between 5.3% and 46.8% of hallucinations depending on category; these misses are unrelated to input length but rise with how far an answer departs from the correct one. Guided by this diagnosis, we propose a two-stage hierarchical detector that first decides whether an answer is hallucinated and then assigns a subtype, raising binary accuracy to 95.6% and subtype accuracy to 56.1%. The results show that type-level evaluation exposes detector weaknesses that aggregate metrics conceal. ## Same Error, Different Function: The Optimizer as an Implicit Prior in Financial Time Series DevFeed: [Same Error, Different Function: The Optimizer as an Implicit Prior in Financial Time Series]() Original publisher: [Read original article]() Author: Federico Vittorio Cortesi, Giuseppe Iannone, Giulia Crippa, Tomaso Poggio, Pierfrancesco Beneventano Published: 2026-10-30T04:00:00Z Content type: news Language: en Sources: [arXiv: Machine Learning]() Topics: [Time Series](), [Finance](), [Artificial Intelligence](), [AI Models]() Tags: [cs-lg](), [financial](), [forecasting](), [model-evaluation](), [neural-networks](), [optimization](), [portfolios](), [q-fin-cp](), [time-series]() ### AI overview Across 12 architecture-optimizer configurations for S&P 500 volatility forecasting, similar held-out errors masked meaningfully different predictors. In 11 of 12 comparisons, changing the optimizer while keeping initialization and data order fixed produced 1.3-7.8 times more divergence than repeated runs of either optimizer; prediction differences were 5-17% of typical forecast error, while error magnitudes differed by under 0.7%. These differences altered response profiles and temporal dependence, and volatility-ranked portfolios showed up to 58% turnover dispersion at comparable gross Sharpe ratios. The results suggest optimization acts as an inductive bias in this underspecified setting, so evaluation should include functional and decision-level behavior as well as scalar loss. ### Source excerpt arXiv:2603.02620v2 Announce Type: replace Abstract: Neural networks applied to financial time series operate in a regime of underspecification, where predictors achieve comparable out-of-sample error. Using large-scale volatility forecasting for S&P 500 stocks--twelve architecture-optimizer configurations, thirteen seeds each, evaluated on 564,795 held-out observations--we show that comparable test loss conceals measurably different predictors. In eleven of twelve comparisons, models trained from identical initialization and identical data order diverge $1.3$ to $7.8$ times further than repeated runs of either optimizer alone, ruling out initialization noise as the explanation. On a common scale, predictions differ by 5--17\% of the typical forecast error while the models' error magnitudes differ by less than 0.7\% of it. The divergence reshapes nonlinear response profiles, temporal dependence, and ultimately decisions: volatility-ranked portfolios trace a Sharpe-turnover frontier with up to 58\% turnover dispersion at comparable gross Sharpe ratios. In underspecified settings, optimization is a consequential source of inductive bias, and model evaluation must extend beyond scalar loss to functional and decision-level behavior. ## Financial Context-Rich Strategy for Actor-Critic based Agents: A Deep Reinforcement Learning Approach for Stock Trading DevFeed: [Financial Context-Rich Strategy for Actor-Critic based Agents: A Deep Reinforcement Learning Approach for Stock Trading]() Original publisher: [Read original article]() Author: Hrishikesh Singh, Aarti Sharma, Millie Pant Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Reinforcement learning](), [Artificial Intelligence]() Tags: [3](), [abstract](), [actor](), [advanced](), [agents](), [algorithmic-trading](), [approach](), [based](), [context](), [cs-ai](), [deep](), [deep-reinforcement](), [deep-reinforcement-learning](), [financial](), [learning](), [q-fin-cp](), [q-fin-pm](), [reinforcement-learning](), [rich](), [stock](), [stock-trading](), [strategy](), [trading]() ### AI overview The abstract reports that sentiment deep-infusion improved profit maximization and risk-averse trading for agents combining policy optimization with Q-learning (DDPG, TD3, SAC). It contrasts these with policy-optimization-only agents (A2C, PPO), but provides no comparable improvement claim for that group. The paper proposes partial and deep sentiment integration to help trading agents account for context beyond price indicators. This overview is based on the abstract excerpt. ### Source excerpt arXiv:2610.32237v1 Announce Type: new Abstract: In modern financial markets, more than 3/4th of the large volume trade is executed by algorithmic trading based on advanced trading strategies to attain higher precision and speed than traditional human traders. The conventional strategies focused on price-based indicators are inadequate in capturing the market dynamism arising from real-world events such as social sentiments, political instability, or sudden economic shifts. With the rise of Deep Reinforcement Learning (DRL) and data availability, there is a requirement for more efficient algorithms that can incorporate these external factors in trading strategies. DRL has shown potential in simulating real-world scenarios and in simplifying complex decision-making since DRL agents can automate the trading process i.e. executing trades in a manner comparable to human traders. They have shown reliable performance since they can update their policies based on risk and quantitative factors. In this paper, we propose a DRL approach based on partial and deep-infusion of sentimental analysis. It features distinct policy designs of actor-critic-based agents: only policy optimization (A2C & PPO) and policy optimization along with Q-learning principles (DDPG, TD3 & SAC). Our findings reveal, that the agents based on policy optimization combined with Q-learning, significantly and consistently demonstrate improved performance upon sentiment deep-infusion in profit maximization as well as risk-averse trading. ## Lineage-Aware Search Control for LLM-Based Evolutionary Factor Mining DevFeed: [Lineage-Aware Search Control for LLM-Based Evolutionary Factor Mining]() Original publisher: [Read original article]() Author: Xintong Liu, Chenkai Zhang, Yanzhao Ma, Peng Yang Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Data Mining & Modeling](), [Large Language Model](), [Hyperparameter optimization]() Tags: [alpha](), [budget](), [control](), [cs-ai](), [cs-lg](), [data](), [discovery](), [diversity](), [efficiency](), [evolution](), [finance](), [generation](), [large-language-models](), [llm](), [markets](), [mining](), [portfolio](), [q-fin-cp](), [quality](), [quantitative-finance](), [search](), [selection](), [semantic]() ### AI overview LineageEvo uses evolutionary descent as a control unit for LLM-based formulaic alpha factor mining. It guides generation with lineage-specific, operator-level priors combined with global priors, and allocates search budget across lineages to limit domination by a few branches while balancing factor quality and lineage concentration. On CSI 500 and S&P 500, it reports the strongest predictive performance; LSPS yields a more balanced lineage distribution and structurally broader factor pool. The abstract also says the evaluation covers portfolio performance, search efficiency, and lineage- and expression-level diversity, but provides no detailed metrics. ### Source excerpt arXiv:2610.29663v1 Announce Type: cross Abstract: Formulaic alpha factor mining searches for symbolic expressions that transform market data into predictive signals. Recent LLM-based evolutionary factor mining methods use large language models as semantic variation operators. Under the conventional LLM-EA paradigm, stochastic variation and fitness-based selection can leave the search weakly directed and concentrated around a narrow set of structurally similar factors. We argue that these limitations arise because current methods mainly organize search at the individual and population levels, leaving lineage-level control underused. A lineage connects successive factor variants through evolutionary descent. We propose LineageEvo, which uses lineage in two coupled forms of search control: (1) lineage-conditioned LLM variation, which distills Lineage-Scoped Search Experience into operator-specific priors and fuses them with global operator priors, providing persistent, lineage-relevant guidance for LLM generation; and (2) lineage-stratified search allocation, which balances factor quality against lineage concentration by distributing the search budget across lineages during evolutionary search. Together, these mechanisms make semantic variation more direction aware and prevent branches from dominating the evolution, thereby promoting the discovery of predictive and diverse factors. We evaluate LineageEvo on CSI 500 and S&P 500 in terms of factor predictiveness, portfolio performance, search efficiency, and lineage- and expression-level diversity. LineageEvo achieves the strongest predictive performance on both markets, while LSPS produces a more balanced lineage distribution and a structurally broader factor pool. The results support lineage as a useful intermediate unit for controlling LLM-based evolutionary factor search.