OntoProbe English Probe Set v1.0 (basic engine output)

OntoProbe is a probe generation engine for the mechanistic study of English language models, plus the structured probe sets that engine produces. This release is the **basic engine output** — the smallest configuration that spans the full 16 vocabulary topics × 16 grammar functions ontology, level-graded to CEFR A1–C2. ### Contents - **32 probe set files** — 16 vocabulary topics × 12 probes + 16 grammar functions × 5 probes = **272 probe sentences** total. - **Top-level ontology schema** (16 topic and 16 function definitions). - **The probe generation engine** itself (`engine/build_probes.py`) — pure Python standard library, no external dependencies. - **A bundled source pool** of 478 vocabulary entries and 160 grammar entries, sufficient to operate the engine and pull up to N=30 vocab / N=10 grammar per cell. The full compiled ontology is not part of this release. ### Intended use Mechanistic interpretability experiments — head specialisation, causal tracing, activation patching, cross-prompt invariance, layer-wise complexity probing, and (with planned future Japanese / Korean releases) cross-language head correspondence studies. ### What is and is not included The probe sets and the engine are released. The full compiled ontology behind the project (~6,700 vocabulary items, ~1,200 grammar patterns with multilingual annotation, subcategory mappings, level distribution, and example pairings) is **not** in this release; the bundled pool is a curated slice of it. ### License Creative Commons Attribution 4.0 International (CC BY 4.0). ### Project Project site: https://ontoprobe.riverbottega.com Maintainer: River Bottega — [e-mail]