pith. sign in

arxiv: 2511.08947 · v5 · submitted 2025-11-12 · 💻 cs.AI

AlphaCast: A Human Wisdom-LLM Intelligence Co-Reasoning Framework for Interactive Time Series Forecasting

Xiaohan Zhang , Tian Gao , Mingyue Cheng , Bokai Pan , Ze Guo , Yaguo Liu , Xiaoyu Tao , Qi Liu This is my paper

Pith reviewed 2026-05-17 23:04 UTC · model grok-4.3

classification 💻 cs.AI
keywords time series forecastinglarge language modelsagentic reasoningiterative forecastingmulti-stage workflowtraining-free modelshuman-AI co-reasoning
0
0 comments X

The pith

Training-free LLMs outperform baselines in time series forecasting by organizing reasoning into a multi-stage expert workflow.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Existing forecasting methods treat prediction as a static single-pass regression. Human experts instead combine temporal features, domain knowledge, case references, and context through continuous iterative refinement. AlphaCast reformulates the task as a multi-stage process of context preparation, reasoning-based generation, and reflective evaluation. It equips training-free LLMs with a lightweight toolkit of features, knowledge base, case library, and contextual pool to support this autonomous interaction. Experiments across benchmarks show the approach generally exceeds representative baselines.

Core claim

AlphaCast enables accurate time series forecasting with training-free large language models by reformulating forecasting as an expert-like process organized into a multi-stage workflow involving context preparation, reasoning-based generation, and reflective evaluation. This transforms forecasting from a single-pass output into a multi-turn, autonomous interaction process. A lightweight toolkit comprising a feature set, a knowledge base, a case library, and a contextual pool provides external support for diverse perspectives in LLM-based reasoning.

What carries the argument

The multi-stage workflow of context preparation, reasoning-based generation, and reflective evaluation, supported by a toolkit of feature set, knowledge base, case library, and contextual pool.

If this is right

  • Time series forecasting shifts from static single-pass regression to dynamic multi-turn refinement.
  • LLMs integrate temporal features, domain knowledge, case references, and supplementary context continuously.
  • Performance exceeds that of representative baselines across multiple benchmarks.
  • Real-world decision-making applications gain from more human-like autonomous forecasting.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same structured toolkit approach could extend to other sequential prediction tasks that benefit from expert-style iteration.
  • Lowering the need for task-specific training may make high-quality forecasting more accessible using general-purpose LLMs.
  • Hybrid setups combining this LLM workflow with occasional human input could further improve reliability in high-stakes domains.

Load-bearing premise

Providing the toolkit of features, knowledge base, case library, and contextual pool is sufficient for training-free LLMs to perform reliable expert-like iterative reasoning on time series data.

What would settle it

A direct comparison on the same benchmarks where AlphaCast without the reflective evaluation stage performs no better than single-pass LLM prompting would show the multi-stage process adds no value.

Figures

Figures reproduced from arXiv: 2511.08947 by Bokai Pan, Mingyue Cheng, Qi Liu, Tian Gao, Xiaohan Zhang, Xiaoyu Tao, Yaguo Liu, Ze Guo.

Figure 1
Figure 1. Figure 1: Illustration of the motivation behind CastMind: shifting forecasting from static, single-pass regression to human-like, inter￾active, multi-turn reasoning. these methods typically treat forecasting as a static, single￾pass regression task, generating predictions based solely on historical observations (Brockwell & Davis, 2002). This approach neglects reasoning, interaction, and iterative re￾finement, which… view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of CastMind. The three-stage process (Investigator, Generator, Reflector) bridges human wisdom and LLM intelligence, simulating expert cognition to transform forecasting into an interactive, multi-turn reasoning process. 3. The Proposed CastMind In this section, we introduce CastMind, which reformulates forecasting as a multi-stage process that simulates human￾like reasoning to prepare… view at source ↗
Figure 3
Figure 3. Figure 3: Component-wise ablation analysis of the forecasting context. The chart illustrates the sensitivity of model performance to the removal of specific information. 4.2. Main Results [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of forecasting performance across different LLM backbones on multiple datasets. 2 4 6 8 10 12 14 Number of Clusters 10 11 12 13 14 MAE (BE & DE) BE DE ETTh ETTm 1.0 1.5 2.0 2.5 MAE (ETTh & ETTm) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Impact of the number of clusters. The optimal granularity balances matching specificity with consensus stability. as PJM and Sunny Power, highlighting the critical role of explicit temporal identifiers in capturing seasonality that numerical sequences alone cannot convey. Similarly, the ex￾clusion of exogenous variables and static data profile leads to substantial error increases on tasks like ETTm and NP,… view at source ↗
Figure 6
Figure 6. Figure 6: Feature-level and model-level case analysis of CastMind across datasets. The heatmaps report the relative usage frequency (%) of temporal features and candidate forecasting models, highlighting a preference for statistical features and deep learning models. Windy Power 0 40 80 120 160 200 0 20 40 60 80 100 DE 25 30 35 40 45 50 0 5 10 15 20 25 Cased-based CastMind Ground Truth Cased-based CastMind Ground Tr… view at source ↗
Figure 7
Figure 7. Figure 7: Case-study comparisons on four datasets (ETTm, MOPEX, Windy Power, and DE), where CastMind produces predictions that track the ground truth more closely than the cased-based prediction, especially under periodic or highly volatile patterns. For instance, on the Windy Power, the error increases drasti￾cally compared to the holistic approach. This deterioration suggests that forcing the model to fragment the… view at source ↗
Figure 8
Figure 8. Figure 8: Prediction performance of different reflection mecha￾nisms across datasets. To evaluate the impact of the reflection strategy on rea￾soning and forecasting quality, we compared our standard full model with an enhanced reflection variant. The lat￾ter retains the same underlying backbone LLM but intro￾duces a more complex and iterative reflection process, along with secondary corrections based on historical … view at source ↗
read the original abstract

Time series forecasting plays a crucial role in decision-making across many real-world applications. Despite substantial progress, most existing methods still treat forecasting as a static, single-pass regression problem. In contrast, human experts form predictions through iterative reasoning that integrates temporal features, domain knowledge, case-based references, and supplementary context, with continuous refinement. In this work, we propose Alphacast, an interaction-driven agentic reasoning framework that enables accurate time series forecasting with training-free large language models. Alphacast reformulates forecasting as an expert-like process and organizes it into a multi-stage workflow involving context preparation, reasoning-based generation, and reflective evaluation, transforming forecasting from a single-pass output into a multi-turn, autonomous interaction process. To support diverse perspectives commonly considered by human experts, we develop a lightweight toolkit comprising a feature set, a knowledge base, a case library, and a contextual pool that provides external support for LLM-based reasoning. Extensive experiments across multiple benchmarks show that Alphacast generally outperforms representative baselines. Code is available at this repository: https://github.com/echo01-ai/AlphaCast.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces AlphaCast, a training-free LLM-based agentic framework for time series forecasting that reformulates the task as an iterative, multi-stage expert-like reasoning process. The workflow consists of context preparation, reasoning-based generation, and reflective evaluation, supported by a lightweight toolkit of temporal features, domain knowledge base, case library, and contextual pool. The central claim is that this setup enables reliable interactive forecasting and that extensive experiments across multiple benchmarks show AlphaCast generally outperforming representative baselines, with code released at a public repository.

Significance. If the performance gains prove robust and reproducible, the work could meaningfully advance time series forecasting by showing how LLMs can be structured for iterative, knowledge-augmented reasoning without training or fine-tuning. This approach integrates human-expert elements such as case-based references and reflective refinement, which may improve interpretability and adaptability in applied settings. The public code release is a clear strength that aids verification.

major comments (2)
  1. Abstract: The claim that 'extensive experiments across multiple benchmarks show that Alphacast generally outperforms representative baselines' is stated without any quantitative metrics, baseline names, dataset details, or error analysis. Because the central contribution is empirical, this omission leaves the primary claim only partially supported and requires a self-contained results summary to allow assessment of effect sizes and consistency.
  2. Methodology section (toolkit and workflow description): The sufficiency of the feature set, knowledge base, case library, and contextual pool for constraining LLM hallucinations and ensuring consistent numerical reasoning is asserted but not demonstrated at the implementation level. Without concrete prompt templates, retrieval algorithms, output validation steps, or ablation results showing each component's contribution, it remains unclear whether the described lightweight toolkit actually delivers reliable expert-like iterative forecasts, which is load-bearing for the outperformance claim.
minor comments (2)
  1. Clarify the exact retrieval and update mechanisms for the case library and contextual pool, including any similarity metrics or update rules, to improve reproducibility.
  2. In the experiments section, report per-dataset breakdowns and statistical significance tests alongside aggregate 'generally outperforms' statements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major comment point by point below, indicating where revisions will be made to strengthen the presentation of our empirical claims and implementation details.

read point-by-point responses

  1. Referee: Abstract: The claim that 'extensive experiments across multiple benchmarks show that Alphacast generally outperforms representative baselines' is stated without any quantitative metrics, baseline names, dataset details, or error analysis. Because the central contribution is empirical, this omission leaves the primary claim only partially supported and requires a self-contained results summary to allow assessment of effect sizes and consistency.

    Authors: We agree that the abstract would be strengthened by including a concise quantitative summary of the results. In the revised version, we will expand the abstract to report specific metrics such as average MAE and RMSE improvements, name key baselines (e.g., ARIMA, LSTM, Informer, and other representative models), reference the primary datasets and benchmarks used, and note the consistency of outperformance across experiments. This will make the central empirical claim self-contained while remaining within abstract length constraints. revision: yes

  2. Referee: Methodology section (toolkit and workflow description): The sufficiency of the feature set, knowledge base, case library, and contextual pool for constraining LLM hallucinations and ensuring consistent numerical reasoning is asserted but not demonstrated at the implementation level. Without concrete prompt templates, retrieval algorithms, output validation steps, or ablation results showing each component's contribution, it remains unclear whether the described lightweight toolkit actually delivers reliable expert-like iterative forecasts, which is load-bearing for the outperformance claim.

    Authors: We acknowledge that the current Methodology section describes the toolkit components at a conceptual level without sufficient implementation specifics. We will revise this section to add concrete prompt templates for each workflow stage, details on the retrieval algorithms employed for the case library and knowledge base, output validation procedures to support numerical consistency, and ablation results that quantify the contribution of each toolkit element (feature set, knowledge base, case library, and contextual pool). These additions will directly address how the components help constrain hallucinations and support reliable iterative forecasting. revision: yes

Circularity Check

0 steps flagged

Empirical framework with external benchmark validation shows no circular derivation

full rationale

The paper presents AlphaCast as an empirical, training-free LLM framework for interactive time series forecasting, structured as a multi-stage workflow (context preparation, reasoning generation, reflective evaluation) supported by a lightweight toolkit of features, knowledge base, case library, and contextual pool. All central claims rest on experimental outperformance across external benchmarks rather than any mathematical derivation, fitted parameter renamed as prediction, or self-referential definition. No equations, uniqueness theorems, or ansatzes are invoked that reduce outputs to inputs by construction; the approach is self-contained against independent data and does not rely on load-bearing self-citations or prior author results for its validity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that LLMs can execute human-like iterative reasoning when supplied with structured external support; no free parameters or new physical entities are introduced in the abstract.

axioms (1)
  • domain assumption Large language models can perform expert-like iterative reasoning on time series when given structured context, knowledge, and cases
    This assumption underpins the entire multi-stage workflow described in the abstract.

pith-pipeline@v0.9.0 · 5513 in / 1161 out tokens · 27167 ms · 2026-05-17T23:04:16.438124+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. CastFlow: Learning Role-Specialized Agentic Workflows for Time Series Forecasting

    cs.LG 2026-04 unverdicted novelty 7.0

    CastFlow introduces a role-specialized agentic workflow with memory retrieval and multi-view toolkit for iterative ensemble time series forecasting, using two-stage SFT+RLVR training on a domain-specific LLM to outper...

  2. GeoDecider: A Coarse-to-Fine Agentic Workflow for Explainable Lithology Classification

    cs.AI 2026-05 unverdicted novelty 6.0

    GeoDecider introduces a coarse-to-fine agentic workflow using LLMs for explainable lithology classification from well logs, combining a base classifier, tool-augmented reasoning, and geological refinement to outperfor...

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages · cited by 2 Pith papers · 15 internal anchors

  1. [1]

    GPT-4 Technical Report

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Chronos: Learning the Language of Time Series

    Ansari, A. F., Stella, L., Turkmen, C., Zhang, X., Mercado, P., Shen, H., Shchur, O., Rangapuram, S. S., Arango, S. P., Kapoor, S., et al. Chronos: Learning the language of time series.arXiv preprint arXiv:2403.07815,

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. Language models are few-shot learners. Advances in neural information processing systems, 33: 1877–1901,

    work page 1901

  4. [4]

    O., Pfister, T., Zheng, Y., Ye, W., and Liu, Y

    Cao, D., Jia, F., Arik, S. O., Pfister, T., Zheng, Y ., Ye, W., and Liu, Y . Tempo: Prompt-based generative pre-trained transformer for time series forecasting.arXiv preprint arXiv:2310.04948,

    work page arXiv

  5. [5]

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    Chen, Y ., Céspedes, N., and Barnaghi, P. A closer look at transformers for time series forecasting: Understanding why they work and where they struggle. InForty-second International Conference on Machine Learning. Comanici, G., Bieber, E., Schaekermann, M., Pasupat, I., Sachdeva, N., Dhillon, I., Blistein, M., Ram, O., Zhang, D., Rosen, E., et al. Gemini...

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [6]

    Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks

    Erdogan, L. E., Lee, N., Kim, S., Moon, S., Furuta, H., Anumanchipalli, G., Keutzer, K., and Gholami, A. Plan- and-act: Improving planning of agents for long-horizon tasks.arXiv preprint arXiv:2503.09572,

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Efficient time series processing for transformers and state-space models through token merging.arXiv preprint arXiv:2405.17951,

    Götz, L., Kollovieh, M., Günnemann, S., and Schwinn, L. Efficient time series processing for transformers and state-space models through token merging.arXiv preprint arXiv:2405.17951,

    work page arXiv

  8. [8]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    Guo, D., Yang, D., Zhang, H., Song, J., Zhang, R., Xu, R., Zhu, Q., Ma, S., Wang, P., Bi, X., et al. Deepseek-r1: In- centivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948,

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    2025 iflytek renewable power forecasting challenge (wind & solar)

    iFLYTEK AI Challenge. 2025 iflytek renewable power forecasting challenge (wind & solar). https://challenge.xfyun.cn/topic/ info?type=renewable-power-forecast& option=ssgy&ch=dwsf259,

    work page 2025

  10. [10]

    Ilbert, R., Odonnat, A., Feofanov, V ., Virmaux, A., Paolo, G., Palpanas, T., and Redko, I

    Accessed: 2026-01. Ilbert, R., Odonnat, A., Feofanov, V ., Virmaux, A., Paolo, G., Palpanas, T., and Redko, I. Samformer: Unlocking the potential of transformers in time series forecasting with sharpness-aware minimization and channel-wise at- tention.arXiv preprint arXiv:2402.10198,

    work page arXiv 2026

  11. [11]

    Time-LLM: Time Series Forecasting by Reprogramming Large Language Models

    Jin, M., Wang, S., Ma, L., Chu, Z., Zhang, J. Y ., Shi, X., Chen, P.-Y ., Liang, Y ., Li, Y .-F., Pan, S., et al. Time-llm: Time series forecasting by reprogramming large language models.arXiv preprint arXiv:2310.01728,

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    Hyperimts: Hypergraph neural network for irregular multivariate time series forecasting.arXiv preprint arXiv:2505.17431,

    Li, B., Luo, Y ., Liu, Z., Zheng, J., Lv, J., and Ma, Q. Hyperimts: Hypergraph neural network for irregular multivariate time series forecasting.arXiv preprint arXiv:2505.17431,

    work page arXiv

  13. [13]

    DeepSeek-V3 Technical Report

    Liu, A., Feng, B., Xue, B., Wang, B., Wu, B., Lu, C., Zhao, C., Deng, C., Zhang, C., Ruan, C., et al. Deepseek- v3 technical report.arXiv preprint arXiv:2412.19437, 2024a. Liu, H., Zhao, Z., Wang, J., Kamarthi, H., and Prakash, B. A. Lstprompt: Large language models as zero-shot time series forecasters by long-short-term prompting.arXiv preprint arXiv:240...

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    Sundial: A Family of Highly Capable Time Series Foundation Models

    Liu, Y ., Qin, G., Shi, Z., Chen, Z., Yang, C., Huang, X., Wang, J., and Long, M. Sundial: A family of highly capable time series foundation models.arXiv preprint arXiv:2502.00816, 2025b. Liu, Z., Cheng, M., Zhao, G., Yang, J., Liu, Q., and Chen, E. Improving time series forecasting via instance-aware post- hoc revision.arXiv preprint arXiv:2505.23583, 20...

    work page internal anchor Pith review Pith/arXiv arXiv

  15. [15]

    WebGPT: Browser-assisted question-answering with human feedback

    Nakano, R., Hilton, J., Balaji, S., Wu, J., Ouyang, L., Kim, C., Hesse, C., Jain, S., Kosaraju, V ., Saunders, W., et al. Webgpt: Browser-assisted question-answering with hu- man feedback.arXiv preprint arXiv:2112.09332,

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    A Time Series is Worth 64 Words: Long-term Forecasting with Transformers

    Nie, Y ., Nguyen, N. H., Sinthong, P., and Kalagnanam, J. A time series is worth 64 words: Long-term forecasting with transformers.arXiv preprint arXiv:2211.14730,

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Tfb: Towards comprehensive and fair benchmarking of time series forecasting methods,

    Qiu, X., Hu, J., Zhou, L., Wu, X., Du, J., Zhang, B., Guo, C., Zhou, A., Jensen, C. S., Sheng, Z., et al. Tfb: Towards comprehensive and fair benchmarking of time series fore- casting methods.arXiv preprint arXiv:2403.20150,

    work page arXiv

  18. [18]

    OpenAI GPT-5 System Card

    10 CastMind: An Interaction-Driven Agentic Reasoning Framework for Cognition-Inspired Time Series Forecasting Singh, A., Fry, A., Perelman, A., Tart, A., Ganesh, A., El-Kishky, A., McLaughlin, A., Low, A., Ostrow, A., Ananthram, A., et al. Openai gpt-5 system card.arXiv preprint arXiv:2601.03267,

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [19]

    Self-Consistency Improves Chain of Thought Reasoning in Language Models

    Wang, X., Wei, J., Schuurmans, D., Le, Q., Chi, E., Narang, S., Chowdhery, A., and Zhou, D. Self-consistency im- proves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171,

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    Deep Time Series Models: A Comprehensive Survey and Benchmark

    Wang, Y ., Qiu, Y ., Chen, P., Zhao, K., Shu, Y ., Rao, Z., Pan, L., Yang, B., and Guo, C. Towards a general time series forecasting model with unified representation and adap- tive transfer. InForty-second International Conference on Machine Learning. Wang, Y ., Wu, H., Dong, J., Liu, Y ., Wang, C., Long, M., and Wang, J. Deep time series models: A compr...

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    ETSformer: Exponential Smoothing Transformers for Time-series Forecasting,

    Woo, G., Liu, C., Sahoo, D., Kumar, A., and Hoi, S. Ets- former: Exponential smoothing transformers for time- series forecasting.arXiv preprint arXiv:2202.01381,

    work page arXiv

  22. [22]

    TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis

    Wu, H., Hu, T., Liu, Y ., Zhou, H., Wang, J., and Long, M. Timesnet: Temporal 2d-variation modeling for general time series analysis.arXiv preprint arXiv:2210.02186,

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    How much can time-related features enhance time series forecasting? arXiv preprint arXiv:2412.01557,

    Zeng, C., Tian, Y ., Zheng, G., and Gao, Y . How much can time-related features enhance time series forecasting? arXiv preprint arXiv:2412.01557,

    work page arXiv

  24. [24]

    Timesense: Making large language models proficient in time-series analysis.arXiv preprint arXiv:2511.06344,

    Zhang, Z., Pei, C., Gao, T., Xie, Z., Hao, Y ., Yu, Z., Xu, L., Xiao, T., Han, J., and Pei, D. Timesense: Making large language models proficient in time-series analysis.arXiv preprint arXiv:2511.06344,

    work page arXiv

  25. [25]

    Least-to-Most Prompting Enables Complex Reasoning in Large Language Models

    Zhou, D., Schärli, N., Hou, L., Wei, J., Scales, N., Wang, X., Schuurmans, D., Cui, C., Bousquet, O., Le, Q., et al. Least-to-most prompting enables complex reasoning in large language models.arXiv preprint arXiv:2205.10625,

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    Method Details A.1

    11 CastMind: An Interaction-Driven Agentic Reasoning Framework for Cognition-Inspired Time Series Forecasting A. Method Details A.1. Feature Overview As detailed in Table 5, the temporal feature set consists of 20 distinct metrics that collectively capture both distributional properties and sequential dependencies of the time series. These features includ...

    work page arXiv 2015