SenTSR-Bench: Thinking with Injected Knowledge for Time-Series Reasoning

Zelin He; Boran Han; Xiyuan Zhang; Shuai Zhang; Haotian Lin; Qi Zhu; Haoyang Fang; Danielle C. Maddix

cs.LGFeb 23, 2026

SenTSR-Bench: Thinking with Injected Knowledge for Time-Series Reasoning

Authors: Zelin He, Boran Han, Xiyuan Zhang, Shuai Zhang, Haotian Lin, Qi Zhu, Haoyang Fang, Danielle C. Maddix,

Organizations: The Pennsylvania State University · AWS AI Labs · Amazon RME

Abstract

Time-series diagnostic reasoning is essential for many applications, yet existing solutions face a persistent gap: general reasoning large language models (GRLMs) possess strong reasoning skills but lack the domain-specific knowledge to understand complex time-series patterns. Conversely, fine-tuned time-series LLMs (TSLMs) understand these patterns but lack the capacity to generalize reasoning for more complicated questions. To bridge this gap, we propose a hybrid knowledge-injection framework that injects TSLM-generated insights directly into GRLM's reasoning trace, thereby achieving strong time-series reasoning with in-domain knowledge. As collecting data for knowledge injection fine-tuning is costly, we further leverage a reinforcement learning-based approach with verifiable rewards (RLVR) to elicit knowledge-rich traces without human supervision, then transfer such an in-domain thinking trace into GRLM for efficient knowledge injection. We further release SenTSR-Bench, a multivariate time-series-based diagnostic reasoning benchmark collected from real-world industrial operations. Across SenTSR-Bench and other public datasets, our method consistently surpasses TSLMs by 9.1%-26.1% and GRLMs by 7.9%-22.4%, delivering robust, context-aware time-series diagnostic insights.

Figures & tables

Figure 1: (a) The newly released SenTSR-Bench benchmark, collected from real-world machine monitoring environments, with multi-stage diagnostic questions. (b) Performance of the proposed framework on SenTSR-Bench , surpassing both stand-alone time-series specialists (TSLM) and general reasoning models (GRLM). (c) Case study illustrating why knowledge injection helps: the specialist captures key time-series patterns but fails to connect them to the correct root cause; the general reasoner shows strong reasoning but overlooks domain-specific critical failure patterns; our method injects the in-domain knowledge from fine-tuned specialist into the reasoner’s reasoning trace , aligning the trace with domain knowledge and producing the correct diagnosis.

Figure 2: Overview of the proposed paradigm. (a) Knowledge injection: given a reasoning question and its time-series, a time–series LM (TSLM) produces grounded analysis snippets that are injected into the reasoning trace of a general frozen reasoning LM (GRLM) to answer diagnostic queries without weight updates. (b) Thinking transfer via RL: We train the TSLM using reinforcement learning with verifiable rewards (RLVR) with an explicit thinking structure to elicit analysis-first thinking traces without human supervision ; at inference, these traces are transferred via injection into the reasoning LM to strengthen temporal grounding for diagnosis.

Benchmark	New Time-Series?	Real- World?	Multi-stage Advancing Questions?	Anno- tation?
TSEvol ( Xie et al., 2024 )	✗	✓	✗	LLM
TS&Language ( Merrill et al., 2024 )	✓	✗	✗	LLM
MTBench ( Chen et al., 2025b )	✓	✓	✗	LLM
Sensor-TSR (Ours)	✓	✓	✓	Human

Table 1: Comparison of time-series diagnostic reasoning benchmarks.

Figure 3: SenTSR-Bench Construction pipeline.

Model	Paradigm	What Happened	How Happened	Suggested Fix	Overall	Improvement vs.
						TSLM	GRLM
TSLM (Qwen-VL-3B)	SFT	0.530 $\pm$ 0.037	0.567 $\pm$ 0.029	0.548 $\pm$ 0.011	0.549 $\pm$ 0.019	—	—
	RL	0.512 $\pm$ 0.038	0.594 $\pm$ 0.019	0.546 $\pm$ 0.009	0.551 $\pm$ 0.014	—	—
GRLM (Claude3.7-Text)	Zero-shot	0.712 $\pm$ 0.019	0.409 $\pm$ 0.033	0.473 $\pm$ 0.024	0.531 $\pm$ 0.011	—	—
	Few-shot	0.691 $\pm$ 0.009	0.561 $\pm$ 0.011	0.509 $\pm$ 0.009	0.587 $\pm$ 0.006	—	+10.5%
TSLM + GRLM	SFT-Injection	0.742 $\pm$ 0.023	0.603 $\pm$ 0.021	0.558 $\pm$ 0.019	0.634 $\pm$ 0.006	+15.5%	+19.4%

Table 2: Reasoning performance on SenTSR-Bench Benchmark (mean

\pm

std). Best per block are bolded . The last two columns report relative gains (in %) for Injection rows vs. the corresponding specialized TSLM and the zero-shot general reasoner (GRLM), respectively.

Model	Paradigm	Causal	Deductive	Inductive	MCQ2	Overall	Improvement vs.
							TSLM	GRLM
TSLM (Qwen-VL-3B)	SFT	0.623 $\pm$ 0.006	0.520 $\pm$ 0.013	0.357 $\pm$ 0.010	0.507 $\pm$ 0.032	0.502 $\pm$ 0.005	—	—
	RL	0.627 $\pm$ 0.016	0.496 $\pm$ 0.014	0.313 $\pm$ 0.023	0.597 $\pm$ 0.031	0.508 $\pm$ 0.006	—	—
GRLM (Qwen3-32B)	Zero-shot	0.507 $\pm$ 0.041	0.473 $\pm$ 0.035	0.623 $\pm$ 0.036	0.407 $\pm$ 0.015	0.502 $\pm$ 0.023	—	—
	Few-shot	0.622 $\pm$ 0.028	0.473 $\pm$ 0.035	0.460 $\pm$ 0.033	0.427 $\pm$ 0.015	0.495 $\pm$ 0.010	—	-1.4%
TSLM+GRLM	SFT-Injection	0.569 $\pm$ 0.035	0.543 $\pm$ 0.013	0.592 $\pm$ 0.031	0.410 $\pm$ 0.036	0.528 $\pm$ 0.008	+5.2%	+5.2%

Table 3: Reasoning performance on TSEvol and TS&Language Benchmark (mean

\pm

std). Best per block are bolded . The last two columns report relative gains (in %) for Injection rows vs. the corresponding specialized TSLM and the zero-shot general reasoner (GRLM), respectively.

Model	Injection Strategy	Inductive	Deductive	Causal	MCQ2	Overall
TSLM (ChatTS-14B)	—	0.812 $\pm$ 0.007	0.597 $\pm$ 0.013	0.732 $\pm$ 0.006	0.590 $\pm$ 0.026	0.683 $\pm$ 0.010
GRLM (Claude3.7-Text)	—	0.763 $\pm$ 0.021	0.612 $\pm$ 0.029	0.645 $\pm$ 0.021	0.640 $\pm$ 0.014	0.665 $\pm$ 0.010
TSLM + GRLM	Intermediate	0.805 $\pm$ 0.026	0.659 $\pm$ 0.022	0.645 $\pm$ 0.010	0.703 $\pm$ 0.037	0.703 $\pm$ 0.006
	Late	0.791 $\pm$ 0.014	0.667 $\pm$ 0.011	0.703 $\pm$ 0.019	0.680 $\pm$ 0.022	0.710 $\pm$ 0.003
	Early	0.824 $\pm$ 0.019	0.643 $\pm$ 0.011	0.703 $\pm$ 0.019	0.690 $\pm$ 0.016	0.715 $\pm$ 0.003
TSLM (ChatTS-14B)	—	0.812 $\pm$ 0.007	0.597 $\pm$ 0.013	0.732 $\pm$ 0.006	0.590 $\pm$ 0.026	0.683 $\pm$ 0.010

Table 4: Performance with different injection strategy on TSEvol and TS&Language Benchmark (mean

\pm

std). Best results are bolded .

Figure 4: Comparison of baseline (zero-shot) reasoning, knowledge prompting, and knowledge injection. (a) SenTSR-Bench Benchmark with Qwen-VL-3B (RL) as the TSLM. (b) TSEvol and TS&Language Benchmarks with Qwen-VL-3B (RL) as the TSLM. (c) TSEvol and TS&Language Benchmarks with ChatTS-14B as the TSLM. Across all settings, the injection-based method consistently outperforms others.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 5: (a) Performance comparison between (i) the standalone TSLM, (ii) knowledge injection where the GRLM receives only the TSLM textual summary (Injection w/o TS), and (iii) full knowledge injection where the GRLM receives both the raw time series and the injected summary (Injection w/ TS). (b) Comparison of overall diagnostic accuracy versus inference latency for different methods.

Figure 6: (a) Performance of the TSLM versus synthetic training data diversity, measured by varying the proportion of seed-generated synthetic data used during training. (b) Comparison of reward trajectories for four RL objectives (GRPO, DAPO, GSPO, CISPO) used to train the TSLM.

Figure 7: Case study on knowledge injection versus standalone baselines. The TSLM correctly detects rising vibration and stable temperature but hallucinates a joint increase in both, yielding an incorrect diagnosis. The GRLM similarly misreads the series, assuming a late temperature rise. Our method leverages the TSLM’s accurate signal interpretation while correcting its reasoning flaw, producing the correct final diagnosis.

Figure 8: Case study on knowledge prompting versus knowledge injection. In the prompting-based approach, the GRLM reasons largely in isolation, referencing the TSLM ’s analysis only at the end for validation, leading to partial use of domain knowledge. In contrast, the injection-based approach integrates the TSLM ’s discussion directly into the reasoning flow, enabling joint exploration and narrowing of hypotheses, and resulting in a correct, well-grounded diagnosis.

Time Series Reasoning

Time series reasoning studies how models interpret temporal patterns, answer questions about sequences, and use evidence to make forecasts.

See all 44 papers

Adaptive Time Series Reasoning via Segment Selection
Feb 20, 2026Shvat Messica, Jiawen Zhang, Kevin Li +2Time Series Reasoning Time Series QA
Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA · The Hong Kong University of Science and Technology (Guangzhou) · Massachusetts Institute of Technology +1
TimeThink: Eliciting Compositional Reasoning in Timeseries Large Language Models
Sep 11, 2026Sudarshan Regmi, Arvind Pillai, Yu Yvonne Wu +7Time Series Reasoning Time Series QA
Dartmouth College
From Recognition to Understanding: Unlocking Cognitive Time Series Reasoning with LLMs
Jun 20, 2026Xin Qiu, Junlong Tong, Yao Zhang +3Time Series Reasoning Time Series QA
Institute of Digital Twin, Eastern Institute of Technology, Ningbo · Zhejiang University · Munich Center for Machine Learning, LMU

Feb 20, 2026J/K move · Enter open · S save

Adaptive Time Series Reasoning via Segment Selection

Shvat Messica, Jiawen Zhang, Kevin Li, Theodoros Tsiligkaridis+1

Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA · The Hong Kong University of Science and Technology (Guangzhou) · Massachusetts Institute of Technology+1

Time Series Reasoning Time Series QA RL for Language Model Reasoning Active Sensing

Open paper AskPDF arXiv

Time series reasoning tasks often start with a natural language question and require targeted analysis of a time series. Evidence may span the full series or appear in a few short intervals, so the model must decide what to inspect. Most existing approaches encode the entire time series into a fixed representation before inference, regardless of whether or not the entire sequence is relevant. We introduce ARTIST, which formulates time-series reasoning as a sequential decision problem. ARTIST interleaves reasoning with adaptive temporal segment selection. It adopts a controller-reasoner architecture and uses reinforcement learning to train the controller role to select informative segments and the reasoner role to generate segment-conditioned reasoning traces and final answers. During inference, the model actively acquires task-relevant information instead of relying on a static summary of the full sequence. We use a novel hierarchical policy optimization approach for post-training that allows the model to excel in both segment selection and question-answering behavior. We evaluate ARTIST on six time-series reasoning benchmarks and compare it with large language models, vision-language models, and prior time-series reasoning systems. ARTIST improves average accuracy by 6.46 absolute percentage points over the strongest baseline. The largest gains appear on rare event localization and multi-segment reasoning tasks. Supervised fine-tuning improves performance, and reinforcement learning provides additional gains by optimizing question-adaptive segment selection. These results show that selective data use drives effective time-series reasoning.

Abstract

Time Series Reasoning

Explore similar work

Adaptive Time Series Reasoning via Segment Selection

TimeThink: Eliciting Compositional Reasoning in Timeseries Large Language Models

From Recognition to Understanding: Unlocking Cognitive Time Series Reasoning with LLMs