LFCarry Guide Evidence Smoke Tests - 24 synthetic cases for guide assistants
LFCarry Guide Evidence Smoke Tests is an open set of 24 synthetic cases for developers evaluating game-guide assistants. Each case pairs a short evidence passage with a claim and an expected label: SUPPORTED, CONTRADICTED or INSUFFICIENT.
The 8 scenario groups cover version changes, platform scope, dates, event boundaries, denominators, prerequisites, missing values and conflicting sources. The downloadable JSONL fixtures include explanations. A dependency-free Node.js evaluator reports coverage, full-set accuracy and a confusion matrix, while rejecting duplicate IDs and invalid labels.
Use the set as a small regression check before testing broader guide workflows. Every scenario uses invented games and facts. AI assistance was used to author the materials; independent human review and actual model-performance measurements are pending. LFCarry maintains the MIT-licensed release.