interwhen: A Generalizable Framework for Verifiable Reasoning with Test-time Monitors
This addresses the need for reliable reasoning in high-stakes domains like law and finance, offering a generalizable solution that is not incremental but introduces a novel approach to verification.
The paper tackles the problem of verifying reasoning models at test-time by proposing interwhen, a framework that verifies reasoning traces as-is using meta-prompting to identify verifiable properties and prompt custom formats. It achieves state-of-the-art results with no accuracy loss in self-verification and a 10 percentage point accuracy improvement with 100% soundness and 4x efficiency in external verification.
We present a test-time verification framework, interwhen, that ensures that the output of a reasoning model is valid wrt. a given set of verifiers. Verified reasoning is an important goal in high-stakes scenarios such as deploying agents in the physical world or in domains such as law and finance. However, current techniques either rely on the generate-test paradigm that verifies only after the final answer is produced, or verify partial output through a step-extraction paradigm where the task execution is externally broken down into structured steps. The former is inefficient while the latter artificially restricts a model's problem solving strategies. Instead, we propose to verify a model's reasoning trace as-is, taking full advantage of a model's reasoning capabilities while verifying and steering the model's output only when needed. The key idea is meta-prompting, identifying the verifiable properties that any partial solution should satisfy and then prompting the model to follow a custom format in its trace such that partial outputs can be easily parsed and checked. We consider both self-verification and external verification and find that interwhen provides a useful abstraction to provide feedback and steer reasoning models in each case. Using self-verification, interwhen obtains state-of-the-art results on early stopping reasoning models, without any loss in accuracy. Using external verifiers, interwhen obtains 10 p.p. improvement in accuracy over test-time scaling methods, while ensuring 100% soundness and being 4x more efficient. The code for interwhen is available at https://github.com/microsoft/interwhen