Repo for "Large Language Model Reasoning Failures"

This is a sample code repository of the power transformer's health state (index) analysis or prediction by the regression model for experiment and learning purposes.

A systems-thinking essay that explains why failure rarely happens suddenly. It shows how slow drift, accumulating pressure, and weakening buffers push systems toward collapse long before outcomes change, and why prediction-focused analytics miss the most important phase of failure.

Sang-Wook's Development and Testing (SWDT)

A systems-thinking essay that reframes failure as a gradual transition rather than a discrete outcome. It explains how pressure accumulation, weakening buffers, and hidden instability precede visible collapse, and why prediction-based models arrive too late to prevent failure in human-centered systems.

Delta-MFP: counterfactual-replay failure diagnosis for local tool-use agents (FAGEN @ ICML 2026)

Giving Every Modality a Voice in Microservice Failure Diagnosis via Multimodal Adaptive Optimization

Sang-Wook's Library (SWL)

Automated AI-powered root cause analysis for GitHub Actions workflow failures.

[IJCNN'24] Official repository of "FashionFail: Addressing Failure Cases in Fashion Object Detection and Segmentation".

Your Playwright results, kept and explained — self-hosted run history, failure clustering, flaky scoring, locator healing.

This repository hosts the project reports for the 'Applied FEA' course, which delves into the application of Finite Element Analysis (FEA) to a wide range of problems. It covers a variety of analyses, such as static, pseudo-static, dynamic, modal, buckling, contact, heat transfer, thermal stress, and thermal shock, all conducted using Ansys.

Failure Compiler for AI agents: turn failed outputs into replayable regression cases.

Hashin criteria predict failure modes in composites based on stresses in fiber and matrix. While effective, implementing them in numerical simulations poses challenges. Abaqus addresses this but is limited to 2D plane stress. A VUMAT subroutine extends Abaqus to simulate 3D damage progression using the energy method for static and dynamic problems.

FAILURE is a Matlab code to simulate the propagation of uncertainties in a plane stress problem subjected to uncertainties.

Adversarial evaluation framework for embodied and agentic AI — failure-first methodology, jailbreak corpus, VLA red-teaming, and policy research.

Open-source failure intelligence: structured startup postmortems, grounded research, and evidence-gated publishing.

A graph-based failure modeling and deterministic detection system for LLM agent runtimes.

MedFailBench exposes critical safety boundary failures in medical AI through clinician built, reproducible open source evaluation.

Research prototype for trace-based observability and failure analysis in retrieval-augmented generation.