Forethought: Verifiable Reasoning from Neurosymbolic Primitive Programming

2026-07-04

AI Summary

Current agent reasoning relies on black-box large models employing chain-of-thought prompting, which suffers from unverifiability, poor auditability, and high computational costs. This work proposes the first model-agnostic, auditable neuro-symbolic reasoning framework that decouples reasoning into explicit programs: by composing symbolic and neural primitives through a domain-specific language, it generates concrete reasoning programs that can be verified and modified prior to deployment. This approach substantially reduces reliance on extensive post-training, boosting base model accuracy by approximately 30% across five benchmarks—enabling small models to match the performance of state-of-the-art large models—while simultaneously reducing inference costs by three orders of magnitude.

Abstract

Current agentic workflows usually involve decomposing user requests into sequences of tool calls with correctly resolved parameters, the results of which are processed through reasoning traces in the language model's context window. The prevailing route to improve such reasoning is test-time scaling, which trains models to search over long chains of thought; but the resulting capability is entangled in model weights, is not verifiable step-by-step, and is costly at inference. We present Forethought, a neurosymbolic reasoning system that instead treats reasoning as an explicit, verifiable program, that builds from a library of symbolic and neural primitives which are composed through a domain-specific language. The result are reasoning programs, which are concrete representations of the model's work, and as such can be inspected and modified before deployment. Instantiated as a tool-calling execution kernel and evaluated across five benchmarks, Forethought improves base-model accuracy by about 30% relative and outperforms vanilla prompting, reinforcement learning scaffolds, and prompt-evolution methods, enabling small models to match or exceed frontier models capabilities. In a direct comparison, a non-reasoning model augmented with Forethought competes with a dedicated reasoning model while requiring roughly three orders of magnitude less post-training investment, and remains model-agnostic and auditable.

Problem

• verifiable reasoning
• neurosymbolic programming
• tool-calling workflows
• reasoning trace
• model-agnostic reasoning

Innovation

• neurosymbolic reasoning
• verifiable programs
• primitive composition
• domain-specific language
• tool-calling execution