Pinned Loading

1. bayes-cot-faithfulness bayes-cot-faithfulness Public

   Bayesian causal-mediation framework for measuring whether a language model's chain-of-thought causes its answer, treating the reasoning as a mediator and reporting a posterior over the direct and i…

   Python 1

2. Olist-Bayesian-AB-Testing Olist-Bayesian-AB-Testing Public

   Hierarchical Bayesian A/B testing on the Olist Brazilian e-commerce dataset. DuckDB SQL pipeline feeding PyMC; a difference-in-differences design flips the sign of the naive on-time-delivery result.

   Python 1

3. car-bench-track1-submission car-bench-track1-submission Public

   Track 1 (Open) entry to the CAR-bench Challenge at IJCAI-ECAI 2026. Placed 2nd of 21 teams on the hidden test set with 66.7% Pass^3 and 86.7% Pass@3. Adds a live tool-schema guard, temperature-zero…

   Python

4. egress-receipts egress-receipts Public

   Egress receipts for AI evaluation sandboxes: one signed file per run that an outsider verifies offline. Apart x CeSIA AI Incident Response Sprint 2026, Track 1.

   TypeScript

5. gauntlet gauntlet Public

   Autonomous red-team for AI applications. Fires attack probes at a chat app, scores it against the OWASP LLM Top 10 (2025), then re-runs the same probes with a runtime guard on. Bundled demo is offl…

   TypeScript

6. lm-kbc-2026 lm-kbc-2026 Public

   Entry to the LM-KBC 2026 challenge on knowledge base construction from language models. Scored 0.7060 all-rows macro-F1 on the official test set, leading on five of the six relations. Final standin…

   Python