Research map / Large language models / Agents, code and math
Analytica: Soft Propositional Reasoning for Robust and Scalable LLM-Driven Analysis
Junyan Cheng, Kyle Richardson, Peter Chin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use linear synthesis rule instead of simple logic rule (ablation) | Analytica on Basic Search · Forecasting 736 events (Basic Search grounder) | accuracy (%): 57.61 -> 65.62 | Improved |
| use linear synthesis rule instead of vanilla rule (ablation) | Analytica on Basic Search · Forecasting 736 events (Basic Search grounder) | accuracy (%): 63.18 -> 65.62 | Improved |
| use linear synthesis rule instead of simple logic (ablation) | Analytica on Basic Search · Forecasting 736 events (Basic Search grounder) | variance of hard score: 7.45 -> 6.46 | Improved |
| add Analytica-L on top of Deep Research grounder | Deep Research · Forecasting 736 events (Deep Research grounder) | accuracy (%): 63.04 -> 71.06 | Improved |
| add Analytica-L on top of Jupyter NB grounder | Jupyter NB · Forecasting 736 events (Jupyter NB grounder) | accuracy (%): 61.96 -> 70.11 | Improved |
| add Analytica-L on top of Basic Search | Basic Search · Forecasting 736 events (Basic Search grounder) | accuracy (%): 53.94 -> 65.62 | Improved |
| use simple logic rule on Jupyter NB grounder instead of plain grounder | Jupyter NB · Forecasting 736 events (Jupyter NB grounder) | accuracy (%): 61.96 -> 62.77 | No clear effect |
| use linear rule instead of simple logic rule on Deep Research (ablation) | Analytica on Deep Research · Forecasting 736 events (Deep Research grounder) | accuracy (%): 66.30 -> 71.06 | Improved |
| use linear rule instead of simple logic rule on Jupyter NB (ablation) | Analytica on Jupyter NB · Forecasting 736 events (Jupyter NB grounder) | variance of hard score: 8.65 -> 7.28 | Improved |
| use powerful grounder (Deep Research) instead of Basic Search under Analytica-L (ablation) | Analytica-L · Forecasting 736 events (Deep Research grounder) | accuracy (%): 65.62 -> 71.06 | Improved |
| use Jupyter NB grounder instead of Deep Research under Analytica-L (ablation) | Analytica-L · Forecasting 736 events | API cost (USD): $14.10 -> $1.36 | Improved |
| increase recursion depth from Analytica to Analytica^4 (ablation) | Analytica · Recursive scalability | avg time (s): 5.3 (1.0x) -> 63.5 (12.0x) | Mixed |
Closest papers by meaning
- Martingale Score: An Unsupervised Metric for Bayesian Rationality in LLM Reasoning (NeurIPS 2025 · similarity 0.88)
- STEER-ME: Assessing the Microeconomic Reasoning of Large Language Models (NeurIPS 2025 · similarity 0.87)
- Fisher-R1: Training LLM Agents for Reliable Hypothesis Testing (NeurIPS 2026 · similarity 0.87)
- STEER: Assessing the Economic Rationality of Large Language Models (ICML 2024 · similarity 0.87)
- ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence (NeurIPS 2026 · similarity 0.86)
- EcoGym: Evaluating LLMs for Long-Horizon Plan-and-Execute in Interactive Economies (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
