Long chain-of-thought (CoT) traces provide observable natural-language rationales for mathematical problem solving, but they rarely expose the named definitions and theorems that justify the written solution. We treat each problem–trace pair as carrying a recoverable knowledge footprint: a trace-grounded support set of named definitions and theorems. Inspired by formal proofs, where supporting premises are recorded more explicitly than in free-form CoT, we formulate Mathematical Support Set Extraction (MSSE): given a problem and a completed long CoT trace, recover the canonical support set that supports the written solution. We propose ReasonMap, which attributes this support set to make solutions auditable, reusable, and easier to repair. ReasonMap follows a two-stage pipeline: proof-to-CoT support transfer adapts premise-explicit proof supervision to premise-implicit long CoT traces, and hard-negative support calibration applies preference optimization against plausible-but-wrong support sets. Experiments show that ReasonMap improves support-set extraction over strong baselines and that ReasonMap-generated support hints aid downstream reasoning in both frontier-model and self-generated settings. The GitHub repository is available at https://github.com/kui253/ReasonMap.

Supervised fine-tuning (SFT) data selection is commonly formulated as instance ranking: score each example and retain a top-k subset. However, effective SFT training subsets are often produced through ordered curation recipes, where filtering, mixing, and deduplication operators jointly shape the final data distribution. We formulate this problem as fixed-pool data recipe search: given a raw instruction pool and a library of grounded operators, the goal is to discover an executable recipe that constructs a high-quality selected subset under a limited budget of full SFT evaluations, without generating, rewriting, or augmenting training samples. We introduce AutoSelection, a two-layer solver that decouples fixed-pool materialization based on cached task-, data-, and model-side signals from expensive full evaluation, using warmup probes, realized subset states, local recipe edits, Gaussian-process-assisted ranking, and stagnation-triggered reseeding. Experiments on a 90K instruction pool show that AutoSelection achieves the strongest in-distribution reasoning average across three base models, outperforming full-data training, random recipe search, random top-k, and single-operator selectors. Additional held-out reasoning, instruction-following, graph-reasoning, search-stability, structural-ablation, and 1.5B-to-7B transfer results further show that recipe structure matters beyond individual selection operators. Code is available at https://github.com/w253/AutoSelection.

The architectural industry produces extensive documents, including method statements—expository documents that integrate multi-source data into actionable guidance. Manual drafting however is labor-intensive and time-consuming. This paper introduces ArchiDocGen, a multi-agent framework automating method statement generation. Unlike traditional approaches relying on static templates or single-pass generation, ArchiDocGen decomposes the task into three steps: outline generation, section-based content generation, and polishing, each handled by specialized agents. To provide domain expertise, ArchiDocGen employs a section-based retriever to fetch and synthesize relevant documents from its custom knowledge base. Each section is generated through iterative reasoning of a section-based chain-of-thought (SeCoT) scheme, followed by refinement to meet professional standards. To evaluate the generated method statements, we partner with the industry to establish a multi-dimensional evaluation system by combining automatic and empirical methods. Experiments show that ArchiDocGen achieves 4.38 ContentScore, excelling in specialization, completeness, organization, and clarity. Additionally, a web-based application for ArchiDocGen is developed and deployed with industry partners.

This paper explores the application of Variational Autoencoders (VAE) in text generation, focusing on overcoming challenges like posterior collapse and the limitations of simplistic prior distributions. We investigate a transition from VAE to text autoencoders (AE), which model a compact latent space and preserves the capability of the language model itself. Our method involves layer-wise latent vectors regularized by orthogonal constraints to encourage distinct semantic spaces. In particular, we estimate an empirical prior online from the learned latent vectors to support sampling during generation like VAE. Experimental results on standard benchmarks demonstrate that the autoencoders generate higher quality and more diverse text than the VAE-based Transformer baselines, offering an effective alternative for generative language modeling.