G.H.O.S.T. is an open-source experiment toolkit for probing the geometric and holographic structure of language model activations. Developed as part of the GCA/HC framework, it implements five experiment modules: - Destructive interference: partial removal of the assistant axis across layers and alpha values- Directional null comparison: random vs. assistant axis vs. role-specific vectors- Pairwise interference: additive steering with pairs of role vectors (holographic memory Property #9)- Weight-space probes: static analysis projecting vectors through k_proj, gate_proj, and other weight matrices- Interactive replay: terminal viewer for saved experiment results (no GPU needed) The toolkit supports Gemma 2 27B, Qwen 3 32B, and Llama 3.3 70B. It includes pre-recorded results for interactive replay, a CLI with presets, a text-based UI, checkpoint/resume, and tmux session wrapping. Role vectors from Lu et al. (arXiv:2601.10387) can be downloaded automatically from HuggingFace at first use. Download Location & Source code: https://github.com/tervehuman/ghost-ts License: MIT