01 DAILY RESEARCH SIGNALS / 2026-10-19
A curated stream of quantitative research, open-source projects, and AI engineering advances—screened, ranked, and structured for faster research decisions.
All sources checked. Pipeline operating normally.
02 CURATED INDEX
Browse by content type or search titles and tags.
INDEXED 15
SELECTED SIGNALS 03
This GitHub project ships a library, a proxy, and an MCP server that compress tool outputs, logs, files, and RAG chunks before they reach the LLM. Its novelty is compressing at the prompt entry point rather than inside the model, claiming 20% fewer tokens for coding agents and 60-95% fewer for JSON with the same answers.
ComfyUI is a diffusion model GUI, API, and backend built around a graph/node interface, surfaced in GitHub's daily trending list. Its value lies in decomposing diffusion workflows into composable nodes, combining visual operation with an API and local inference backend for building and reusing pipelines.
A JavaScript repository named withmarbleapp/os-taxonomy appears on GitHub daily trending, offering an operating system taxonomy. Its novelty lies in publishing such a classification structure openly for reuse and discussion; its usefulness is limited to contexts that need the framework itself.
SELECTED SIGNALS 08
The study proposes a multivariate state-dependent Hawkes process to model cross-product liquidity dynamics between quarter-hourly and hourly contracts in the Dutch EPEX SPOT intraday continuous market, distinguishing normal and stressed regimes. Its novelty lies in quantifying how cross-product excitation weakens in direction and accelerates in speed under stress, helping to understand hidden liquidity dependencies among overlapping contracts.
Develops a framework to identify and quantify arbitrage across logically related prediction-market contracts, using a convex-hull condition for arbitrage existence and integer programming for optimal portfolios, with order-book replay on Polymarket to quantify net profits and persistence.
Proposes the AOI multi-agent framework that formulates autonomous operations as a trajectory learning problem under security constraints. Its novelty lies in converting failed trajectories into supervision signals via a failure trajectory closed-loop evolver, with cumulative gains validated on AIOpsLab.
The study introduces OrderFlowDiT, a conditional diffusion Transformer for jointly forecasting unnormalized bid and ask volume-at-price fields over the next five minutes, with an unbalanced optimal transport (UOT) auxiliary loss to explicitly account for price placement and total volume. It also develops a full-field evaluation framework combining a transport cost sensitive to price placement and total volume with proper scoring rules and calibration diagnostics.
The study addresses hindsight bias in financial news language models, which often train on text overlapping the test period, and proposes NoLBERT—pretrained only on 1976-1995 text—to construct news topics. It compares topic extractors on 1.26 million articles from 2012 to 2024 for predicting U.S. stock returns, systematically testing the feasibility and stability of hindsight-free topic extraction and highlighting the role of the attention-shock measurement window.
The study jointly models execution horizon choice and schedule shape under concave power-law impact and Hurst-scaled price fluctuations, proposing Sharpe ratio maximization to determine the optimal window and deriving an analytical schedule via convex programming.
CastFlow addresses two key challenges in LLM-based agentic time series forecasting—acquiring forecast-relevant evidence through adaptive tool interactions and translating evidence into accurate numerical predictions—by proposing a role-specialized framework: a frozen LLM, guided by experience memory, orchestrates a multi-view toolkit for evidence acquisition, while a lightweight specialized LLM performs evidence-grounded temporal reasoning to refine an ensemble forecasting pr
Proposes RePlanMem-VLA, which uses large disagreement between action predictions during policy replanning as an online memory-write signal to selectively store key visual frames, addressing the need for historical visual information in long-horizon VLA policies without expensive and redundant full history storage.
SELECTED SIGNALS 02
This item records that Import AI author Jack Clark said in his newsletter that he told the International Monetary Fund (IMF) that 'the singularity has begun.' Its novelty lies in bringing a general-AI narrative into international financial governance; its usefulness is as a public example of AI–macro policy dialogue, not as verifiable technical or economic evidence.
This is an aggregator-style link summary whose value lies in collecting recent quant community discussions and using a Jeff Yass quote to highlight a core risk: mistaking a losing strategy for a winning one. For quant researchers, it serves as a pointer to community trends and a prompt to question strategy evaluation assumptions.
SELECTED SIGNALS 02
This content introduces Pi Durable as a building block in the Pi ecosystem for always-on agent applications, enabling recovery after interruptions via checkpoints, conversations, and storage, and demonstrates a local llama.cpp demo with Qwen3.5-0.8B.
Pi Durable, presented by Hugging Face, is a TypeScript library whose stated novelty is checkpointing every task and tool call (SQLite / JSONL) so an agent resumes where it left off after a process restart, instead of leaving a half-finished tool call. It targets always-on agent builds and is described as usable fully locally with Qwen3.5-0.8B on llama.cpp.