Benchmarking large language models for supply chain risk identification: an extended evaluation within the LARD-SC framework
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Abstract Operational resilience in modern global supply chains depends on timely and accurate identification of emerging risks. While daily news has become a primary source for such insights, the sheer volume and unstructured nature of these data pose significant analytical challenges, requiring advanced tools to extract relevant and actionable information. This paper introduces an extended evaluation of the LARD-SC framework, a service-oriented architecture for supply chain risk management, by benchmarking five diverse variants of the large language model (LLM) in their capacity to detect, classify, and interpret risks. Drawing on a curated set of 120 real-world news articles on Apple’s Tier 1 suppliers, we adopt a standardized, prompt-based assessment to compare GPT-3.5 turbo, GPT-4o, GPT-4o mini, Claude 3.5 Sonnet, and Claude 3.5 Haiku. Using expert-reviewed metrics, namely the Risk Validation Rate (RVR), Potential Risk Rate (PRR), and False Identification Rate (FIR), we derive a comprehensive Relative Performance Index (RPI) for comparison. Our analysis confirms that advanced GPT-4o variants produce the most consistent accurate risk identifications, achieving higher proportions of validated outcomes while minimizing false positives. Through these results, we highlight the significant promise of LLM-driven analytics for early risk detection in complex supply chains, along with practical considerations such as the influence of prompt engineering, interpretability demands, and the impact of data availability. The findings offer a blueprint for organizations seeking to improve resilience by systematically harnessing the capabilities of LLM within service-oriented risk management ecosystems.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it