mlsanityData to 7 Oct 2026

TEMPO: Temporal Enforcement via Mode-Separated Policy Optimization for Trustworthy LLM Backtesting

arXiv 2605.18843 ↗First posted 13 May 2026cs.LGcited by 0

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Backtesting large language models on historical events requires reasoning exclusively from information available before a specified cutoff date. Yet models routinely leak post-cutoff knowledge from pre-training into their reasoning, inflating apparent accuracy and undermining evaluation validity. Prompt-based constraints fail when suppressed content is causally related to the prediction, and knowledge unlearning cannot address this problem because temporal compliance is instance-specific: the same fact may be legitimate evidence for one cutoff date and a violation for another. Rather than erasing knowledge, the model must learn temporal discipline: selecting evidence conditioned on each instance's cutoff date. We propose TEMPO (Temporal Enforcement via Mode-separated Policy Optimization), which trains this discipline via two contributions: (1) a two-mode reward where a leakage mode drives post-cutoff claims to zero as a hard prerequisite before a performance mode optimizes task performance; and (2) a GRPO-based training pipeline that enables the model to discover temporally valid reasoning strategies. We prove that training monotonically decreases leakage, converges to the leak-free optimum, and improves task performance once compliance is achieved. On three prediction tasks and two models, TEMPO reduces leakage from 2~13% to 0.6~3.7% across all conditions, with task performance improving 6~13% where strong pre-cutoff signals exist and maintained where the prediction task is inherently difficult from valid information alone.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 33 of its 42 references that are in the map. First posted on arXiv 13 May 2026.

Builds on

33 of 42 references.

References · 33

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  4. 2015Trust Region Policy Optimizationcited by 2,503
  5. 2020Extracting Training Data from Large Language Modelscited by 2,144
  6. 2023BloombergGPT: A Large Language Model for Financecited by 923
  7. 2013Stochastic First- and Zeroth-order Methods for Nonconvex Stochastic Programmingcited by 716
  8. 2023Safe RLHF: Safe Reinforcement Learning from Human Feedbackcited by 630
  9. 2016Linear Convergence of Gradient and Proximal-Gradient Methods Under the Polyak-Lojasiewicz Conditioncited by 526
  10. 2026GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimizationcited by 175
  11. 2019On the Theory of Policy Gradient Methods: Optimality, Approximation, and Distribution Shiftcited by 174
  12. 2024ForecastBench: A Dynamic Benchmark of AI Forecasting Capabilitiescited by 104
21 more
  1. 2024Approaching Human-Level Forecasting with Language Modelscited by 102
  2. 2024Wisdom of the Silicon Crowd: LLM Ensemble Prediction Capabilities Rival Human Crowd Accuracycited by 66
  3. 2024Digital Forgetting in Large Language Models: A Survey of Unlearning Methodscited by 38
  4. 2024Erasing Conceptual Knowledge from Language Modelscited by 34
  5. 2025A Survey on Data Contamination for Large Language Modelscited by 33
  6. 2023A Comprehensive Evaluation of Large Language Models on Legal Judgment Predictioncited by 27
  7. 2025Pitfalls in Evaluating Language Model Forecasterscited by 23
  8. 2025BLUR: A Bi-Level Optimization Approach for LLM Unlearningcited by 23
  9. 2025MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problemscited by 21
  10. 2025TruthRL: Incentivizing Truthful LLMs via Reinforcement Learningcited by 20
  11. 2023History Matters: Temporal Knowledge Editing in Large Language Modelcited by 17
  12. 2024Enabling Discriminative Reasoning in LLMs for Legal Judgment Predictioncited by 16
  13. 2026Chaining the Evidence: Robust Reinforcement Learning for Deep Search Agents with Citation-Aware Rubric Rewardscited by 14
  14. 2025Profit Mirage: Revisiting Information Leakage in LLM-based Financial Agentscited by 13
  15. 2025UIPE: Enhancing LLM Unlearning by Removing Knowledge Related to Forgetting Targetscited by 10
  16. 2025Can Prompts Rewind Time for LLMs? Evaluating the Effectiveness of Prompted Knowledge Cutoffscited by 8
  17. 2024Temporal Blind Spots in Large Language Modelscited by 7
  18. 2026When Silence Is Golden: Can LLMs Learn to Abstain in Temporal QA and Beyond?cited by 5
  19. 2026The Impact of Post-training on Data Contaminationcited by 4
  20. 2024Estimating Player Performance in Different Contexts Using Fine-tuned Large Events Modelscited by 3
  21. 2026Constrained Group Relative Policy Optimizationcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Temporal Sampling for Forgotten Reasoning in LLMs
  2. 2026Teaching Large Language Models When Not to Know: Learning Temporal Critique for Ex-Ante Reasoning
  3. 2026TEMPO: Scaling Test-time Training for Large Reasoning Models
  4. 2026Policy of Thoughts: Scaling Test-Time Training for LLM Reasoning via Online Policy Evolution
  5. 2025HAPO: Training Language Models to Reason Concisely via History-Aware Policy Optimization
  6. 2026CoEM: Empowering Long-Context Reasoning with Commit-on-Evidence Memory
  7. 2025Less is More: Improving LLM Reasoning with Minimal Test-Time Intervention
  8. 2026Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning