Aionoscope: Debugging Latent-State Accessibility in Time-Series Representations

2026-07-01

AI Summary

Current evaluations of time series models predominantly focus on forecasting or classification performance, offering limited insight into whether their learned representations preserve fine-grained latent states of interest—such as event timing, phase, and frequency. To address this gap, this work proposes Aionoscope, the first diagnostic framework that systematically disentangles process generation from observation rendering. It leverages Primitive Process Mixtures to produce controllable synthetic data with precise class labels and dense annotations, enabling a unified evaluation of 37 model-adapter systems via a linear probing protocol. Experiments reveal that while most systems adequately recover component presence, they exhibit substantially weaker capability in reconstructing dense process states: the best-performing dense probe achieves a mean masked R² of only 0.689, far below the ideal value of 0.999, highlighting a pronounced disparity between coarse- and fine-grained representational accessibility.

Abstract

Time-series models are often evaluated by what they can forecast or classify, but those scores do not show whether their representations preserve the process state a user may want to inspect: event timing, phase, amplitude, frequency, or regime variables. We introduce Aionoscope, a generator-based diagnostic tool for debugging latent-state accessibility in frozen time-series representations. Aionoscope separates process generation from observation rendering, producing seeded synthetic streams with exact categorical and dense labels across mixture complexity and nuisance variation. We instantiate Aionoscope as Primitive Process Mixtures and evaluate 37 model-plus-adapter systems with a common pooled linear-probe protocol. The main result is a mismatch between coarse and fine-grained accessibility. Most systems make component presence easy to recover, but expose dense process state much less reliably: the highest observed dense-probe row reaches 0.689 mean masked $R^2$, while a dense-feature oracle reaches 0.999. This is the failure mode Aionoscope is designed to surface: a representation can look informative at the level of "what kind of signal is present" while hiding the timing, phase, amplitude, frequency, or regime variables needed for debugging.

Problem

• latent-state accessibility
• time-series representations
• process state
• debugging
• representation evaluation

Innovation

• latent-state accessibility
• time-series representation
• generative diagnostics
• synthetic data probing
• fine-grained process state