Latent chain-of-thought compresses reasoning by replacing visible reasoning traces with continuous hidden-state recurrence, but existing formulations are difficult to optimize with standard on-policy reinforcement learning (RL) and hard to interpret causally. Our key insight is that a single pair of explicit boundary tokens can address both issues at once: discrete entry and exit anchors make the latent block compatible with standard on-policy RL, and the same anchors offer a natural foothold for mechanistic analysis. Motivated by this, we propose SWITCH, a switchable latent reasoning framework. The model emits ‘<swi>‘ to enter latent mode and ‘</swi>‘ to exit. Because the boundaries are ordinary discrete tokens, the GRPO policy ratio is well-defined at every decision point. The same anchors also expose the latent steps to direct probing and causal intervention. We train the model with a visible-to-latent curriculum and a Switch-GRPO objective that propagates gradients through recurrent latent computation. SWITCH consistently outperforms prior hidden-state-recurrence latent reasoning approaches at similar scale. Mechanistic analysis through the boundary tokens further reveals three findings: (i) ‘<swi>‘ is a sharply localised, learned switching policy rather than a stylistic artefact; (ii) the latent step it opens performs problem-specific, causally important computation rather than acting as an inert placeholder; and (iii) that computation is concentrated at a single hidden-state transition on entry. Together, these results show that hidden-state-recurrence latent reasoning is both RL-trainable and open to direct mechanistic analysis, including of how on-policy RL itself improves the model from the inside.

Large language models (LLMs) have shown strong multimodal reasoning abilities, yet their understanding of purely numerical time-series remains limited. Existing approaches mainly focus on forecasting or trend description, providing little insight into the underlying events that have occurred or are unfolding within observed numerical dynamics. We study number-to-event reasoning and decoding, a task that infers interpretable and structured event hypotheses directly from numerical time-series when explicit event information is unavailable at inference time. The task focuses on aligned and plausible event semantics that help explain observed numerical dynamics. To address data scarcity and semantic alignment challenges, we propose a reasoning-aware framework that combines an agent-guided event extractor, a Hawkes-process-based synthetic data generator, and a multi-stage fine-tuning pipeline integrating a time-series encoder with a structured event decoder. The model generates intermediate reasoning steps and outputs constrained event representations. Experiments across multiple real-world domains show that our approach consistently outperforms strong LLM baselines in event-level precision and recall, demonstrating its effectiveness in bridging numerical reasoning and semantic event understanding. The code is available at https://github.com/feng1201/NUM2EVENT_IDEA

Reference-conditioned image generation increasingly relies on visual histories that rarely speak with one voice: one example pins the subject, another fixes a layout, a third only suggests a palette. Current image-conditioned diffusion methods encode these references once, average them into a single vector, and inject it at every denoising step, so conflicting cues collide and the output is faithful to none. We propose the Hierarchical Temporal Guidance Framework (HTGF), a training-free pipeline that reframes multi-reference conditioning as a *temporal routing* problem. A semantic decomposer reads each reference and emits a soft routing distribution over three axes (Subject, Structure, Detail), and a closed-form SNR-sensitivity argument places each axis in the denoising window where it has the most leverage. A short manifold-aware corrector then smooths the trajectory at the stage boundaries where the active condition changes. HTGF adds no training and no architectural change to the diffusion backbone. On three datasets it outperforms strong VLM- and diffusion-based baselines, gaining +11.7 CIS over LaVIT on Movie Poster and FID 38.87 vs. 50.12 for the best baseline under noisy histories, and degrading gracefully down to the zero-reference limit.