# Decoupled Predictive Decision Making: A Shared Conditional Predictor for Agents

DevFeed: [Decoupled Predictive Decision Making: A Shared Conditional Predictor for Agents](<https://devfeed.tech/articles/decoupled-predictive-decision-making-a-shared-conditional-predictor-for-agents-115264.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02075>)

Author: Lang Mo, Xingxuan Zhang, Qi Wang, Mingchao Hao, Hao Yuan, Chunjing Gan, Wang Lu, Peng Cui

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Inference](<https://devfeed.tech/topics/inference.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [model-deployment](<https://devfeed.tech/topics/model-deployment.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [decision-making](<https://devfeed.tech/tags/decision-making.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [structured-data](<https://devfeed.tech/tags/structured-data.md>)

## AI overview

DPDM casts varied agent decisions as conditional prediction: a language model represents current and historical evidence, and a shared pretrained predictor estimates task-defined targets; a decision rule turns predictions into choices. Both models remain frozen at deployment, where observed histories supply task-specific evidence. Using one frozen LimiX-2 predictor across five task categories with Qwen3.5-27B as encoder, the reported mean accuracy across 50 benchmarks is 84.92% with each method's evaluated history budget versus 78.80% for few-shot prompting, and 82.33% with identical histories and queries. Reported gains on ToolPRMBench's BFCL and ToolSandbox are 34.26 and 28.87 percentage points at respective evaluated history budgets. Accuracy rises with more history; on five reasoning-step assessment tasks, balanced accuracy improves by 14.51 points with identical imbalanced histories across four encoders. These findings are limited to the reported evaluations.

## Source excerpt

arXiv:2611.02075v1 Announce Type: new Abstract: Agent decisions depend on empirical relationships that deployment observations reveal and task descriptions alone may leave unresolved. We introduce Decoupled Predictive Decision Making (DPDM), an architecture that formulates heterogeneous agent decisions as conditional prediction with a shared pretrained learner. Language models provide semantic representations of current and historical records, while a general-purpose conditional predictor predicts task-defined targets from the represented evidence. Pretraining supplies reusable inference capability; deployment evidence instantiates problem-specific predictions without updating either model's neural parameters. Tool judgments, candidate preferences, and reasoning assessments share an evidence-and-query interface, with predictions mapped to choices by a decision rule. This architecture establishes complementary sources of agent capability: semantic knowledge acquired through language pretraining and empirical inference acquired through structured-data pretraining. A single frozen LimiX-2 predictor serves all five task categories. With Qwen3.5-27B as the semantic encoder, DPDM raises mean accuracy across 50 benchmarks from 78.80% with few-shot prompting to 84.92% under their respective evaluated history budgets, and reaches 82.33% with identical histories and queries. On ToolPRMBench's BFCL and ToolSandbox tasks, absolute accuracy gains are 34.26% and 28.87%, respectively, at the methods' respective evaluated history budgets. Accuracy improves as history grows. With identical imbalanced histories, mean balanced accuracy improves by an absolute 14.51% across five reasoning-step assessment tasks and four encoders. DPDM makes learning from experience an independently improvable foundation-model capability, creating a shared route for advances in conditional prediction to improve decisions across agent tasks.