Empirical Activations DatasetRaw activation data from the empirical validation of the GCA/HC framework.Companion report: Empirical Validation of the GCA/HC Framework (DOI: 10.5281/zenodo.18734207)Theoretical paper: The LLM Naming Problem (DOI: 10.5281/zenodo.17221959)Experiment toolkit: G.H.O.S.T.## Models| Model | Layers | Hidden dim | Architecture ||-------|--------|------------|-------------|| Gemma 2 27B | 46 | 4608 | Google DeepMind || Qwen 3 32B | 64 | 5120 | Alibaba || Llama 3.3 70B | 80 | 8192 | Meta |## Dataset structure01-empirical-activations/├── t1_t3_multiturn/           §5      Multi-turn drift (Gemma)         11 MB├── t2_phase_transitions/      §4      Phase transitions (Gemma)       363 MB├── directional_null/          §6.6-7  Directional comparison (3 mod)  1.3 GB├── pairwise/                  §7.8    Pairwise interference (Qwen)    384 MB├── pairwise_dense/            §7.8    Dense pairwise (Qwen, extended) 1.8 GB└── weight_probe/              §7.8    Weight-space probes (Qwen)      1.4 GBTotal: ~5.2 GB, ~16,600 files.Each subdirectory contains a README with details on the experiment, file format, and the relevant report sections.## MethodologyAll activation data was generated using the assistant axis vectors from Lu et al. (arXiv:2601.10387). The core technique is partial projection removal:h <- h - alpha * <h, a_hat> * a_hatwhere h is the hidden state, a_hat is the unit assistant axis, and alpha controls ablation strength (0 = no intervention, 1 = full removal).## File formatAll .npy files are NumPy arrays saved with numpy.save(). Load with:pythonimport numpy as npdata = np.load("filename.npy")## LicenseCC BY 4.0