O.W.L. & N.E.W.T. Bench

What is O.W.L. & N.E.W.T. Bench?

Deliberately contaminated LoRA experiment with a 75-question Harry Potter evaluation set and training code that reproduces the leakage demonstration.

Released

2026-09-08

Evaluates

Language & Knowledge, Software & AI Compute, cs.AI

Openness

unknown

Importer

Claire Radar

Review status

ai-name-audit-deferred

Reported scores

0

Source provenance

• Original evidence https://github.com/VSBDev/little-hermione

O.W.L. & N.E.W.T. Bench code and dataset

• Code https://github.com/VSBDev/little-hermione
• Dataset https://huggingface.co/datasets/irioder/littleHermione-benchmark
• Website https://victorsb.com/experiments/littlehermione/

Which sources cite O.W.L. & N.E.W.T. Bench?

• O.W.L. & N.E.W.T. Bench

Related Language & Knowledge, Software & AI Compute, cs.AI benchmarks

• 5-Dialects-BN
• A Large-Scale Dataset and Benchmark
• AARRI-Bench
• ABC-Bench
• Active-SWE
• ActiveSciBench
• AdaMem-Bench
• AdvancedMathBench