Chain of News Digest

Chain of News 02/10/2026

02/10/2026
**Top Story** ContractRL introduces a novel reinforcement‑learning framework that enforces execution contracts on tool‑calling agents, dramatically improving the auditability of automated repairs. By treating schema violations as contract breaches rather than merely error signals, the system can regenerate only the offending fields, keeping the rest of the action intact and preserving provenance. For developers building LLM‑driven assistants that invoke external APIs or code generators, this means fewer costly round‑trips, tighter safety guarantees, and clearer logs for compliance teams. The paper also demonstrates that ContractRL scales to real‑world tool suites without sacrificing performance, positioning it as a practical alternative to heavyweight sandboxing. As enterprises push LLM agents deeper into critical workflows, the ability to audit and correct tool calls on the fly could become a de‑facto standard for responsible AI deployment. SOURCES: [3] **AI Models & Research** The “Microtask Eligibility Gap” study probes when off‑the‑shelf small language models suffice for the peripheral tasks that support a central LLM planner, revealing that task‑specific fine‑tuning often yields diminishing returns for simple micro‑operations. Gradient‑Aligned Pair Selection advances personalized preference optimization by selecting training pairs that align gradient directions with individual user signals, offering developers a more data‑efficient path to tailor LLM outputs without exhaustive labeling. The mathematical transfer analysis shows that LLMs benefit more from shared reasoning approaches than from topical similarity, suggesting that curating training sets around problem‑solving strategies can accelerate the acquisition of robust mathematical abilities. SOURCES: [1], [7], [8] **Developer Tools & Frameworks** K‑Dense BYOK delivers an open‑source research assistant that runs entirely on a user’s hardware, letting scientists plug in any local or hosted model while maintaining a hash‑chained notebook for immutable experiment provenance. This enables teams to comply with strict data‑privacy regimes while still leveraging cutting‑edge LLM capabilities. Rules to Tools (R2T) equips scientific coding agents with executable checks that validate equations, boundary conditions, and output specifications against a library of public scientific standards, turning vague textual prompts into enforceable contracts. Developers can now embed these checks directly into their agent pipelines, catching domain‑specific errors before costly compute is spent. SOURCES: [4], [6] **Industry & Business** Build2SPARQL releases a large‑scale benchmark that translates natural‑language queries into SPARQL for building automation knowledge graphs, accelerating the integration of AI into smart‑infrastructure management. The dataset leverages industry ontologies like Brick and ASHRAE 223P, giving vendors a ready‑to‑use testbed for deploying LLM‑driven query interfaces that can surface actionable insights from complex sensor networks. SOURCES: [2] **Worth Watching** EviGraph proposes a proof‑carrying recommendation engine over temporal public‑service knowledge graphs, distinguishing critical decision requirements from optional evidence to avoid over‑pruning useful suggestions. Its approach could reshape how civic tech platforms surface services, balancing transparency with relevance, and may inspire similar evidence‑aware models in other domains. SOURCES: [5]

Today's Stories

Today's articles

ArXiv cs.AI

EviGraph: Proof-Carrying Selective Recommendation over Temporal Public-Service Knowledge Graphs

Public-service recommendations require evidence that matches the requested service, scope, and date. Yet treating every missing detail as decisive can withhold useful recommendations. We introduce EviGraph, which distinguishes critical decision requirements from information that can remain unresolved. A language agent links these requirements to evidence in a temporal knowledge graph, while a deterministic checker establishes whether a recommendation is supported.

02/10/2026
ArXiv cs.AI

Rules to Tools: Executable Checks for LLM Agents in Scientific Computing

Scientific coding agents receive equations, boundary conditions, and output requirements in writing, then must assess the programs they revise. Rules to Tools (R2T) supplies prepared executable checks of public scientific requirements. Matched SciCode repair groups share written checks, starting programs, model, and budgets; the tool group receives a callable implementation. Across two task-ID cohorts, complete repair is 26/30 with text and 29/30 with the prepared checks.

02/10/2026
ArXiv cs.AI

ContractRL: Shielded Group-Relative Policy Optimization for Auditable Tool-Call Repair

Structured tool calls often fail after only a small number of fields violate a schema or an execution contract. Regenerating the complete object enlarges the action surface and makes repeated repair difficult to audit. We introduce ContractRL, a contract-constrained sequential repair protocol that models verifier-guided JSON repair as a bounded decision process.

02/10/2026
ArXiv cs.AI

Mathematical Transfer in LLMs Follows Reasoning Approach More Than Topic

When selecting mathematical training data for LLMs, a natural organizing principle is topic: probability examples for probability targets. An alternative is reasoning approach: worked solutions that share a solution method with the target, even when the mathematical domain differs. We ask which relation produces greater transfer after fine-tuning.

02/10/2026
ArXiv cs.AI

K-Dense BYOK: An Open-Source AI Research Assistant That Runs Locally and Keeps a Hash-Chained Lab Notebook

K-Dense BYOK (bring your own keys) is a free, open-source AI research assistant for scientists in any field that runs on the researcher's own computer. The researcher supplies access to a model of their choice, hosted or running locally, and the application supplies everything else: a place for the work to run, a layer of scientific scaffolding, and a complete record.

02/10/2026
ArXiv cs.AI

Build2SPARQL: A Large-Scale Text-to-SPARQL Benchmark Dataset for Building Knowledge Graph Querying

Building automation systems are increasingly represented as semantic knowledge graphs (KGs) using ontologies such as Brick and ASHRAE 223P, creating a machine-readable substrate for artificial-intelligence applications. One promising application is translating natural-language questions into SPARQL (text-to-SPARQL), which would let building operators query these graphs through language agents, but progress is limited by the scarcity of large natural-language/SPARQL benchmarks.

02/10/2026
ArXiv cs.AI

Measuring the Microtask Eligibility Gap: When Is an Off-the-Shelf SLM Enough for an Agent Harness?

Agent harnesses increasingly want to run small language models (SLMs) on the microtasks around a frontier large language model (LLM) planner: auto-approving shell commands, writing memory, selecting tools, ranking past turns. We ask whether off-the-shelf SLMs meet practitioner-defined thresholds and, when they fail, why, and whether quantization changes the answer.

02/10/2026
ArXiv cs.AI

Gradient-Aligned Pair Selection for Personalized Preference Optimization

Personalizing large language models (LLMs) requires aligning generation behavior with user-specific preferences rather than aggregate quality. While Direct Preference Optimization (DPO) provides a stable framework for preference learning, its effectiveness in personalized settings critically depends on how preference pairs are selected.

02/10/2026