Paper artifact

Reproduce the TGCM experiments

The artifact follows the figure and table numbering used in the paper. Each experiment includes a focused README, with executable materials being prepared for release.

Browse experiments

Experiment index

Paper-to-artifact mapping

Experiment descriptions are available now. Executable notebooks and supporting files will be enabled with the complete reproduction release.

Figure 4

Blind unknown-K comparison

TGCM, DANet, MossFormer2, DECOMPOSE, OpenAI, and Gemini across K=2 through K=6.

READMENotebookLLM prompt

Tables IV & XII

Zero-shot coverage

Technique-to-kill-chain mapping and grouped coverage counts.

READMENotebookMapping

Tables V & XIII

Zero-shot performance

Five-seed TGCM and DANet evaluation on ATLAS, NODLINK, ProvCon, DARPA TC-E3, and TC-E5.

READMENotebook

Tables VI & XXIII

DARPA robustness

Missing, confusion, and insertion perturbations over the complete DARPA TC-E5 evaluation grid.

READMENotebook

Tables VII & XXI

CAPTure evaluation

Pooled single-host and multi-host evaluation using SFM, Zoomer, and TREC outputs.

READMENotebook

Table XXII

Unknown-K sensitivity

Known-K versus Up-to-6 decoding from identical TGCM forward outputs.

READMENotebook

Table XXV

Embedding sensitivity

Seven topic-encoder families, five seeds, and mixture levels K=2 through K=6.

READMENotebook

Environment

Quick setup

conda env create -f environment.yml
conda env create -f 01_Figure04_Blind_Unknown_K/environment_decompose.yml
conda activate tgcm-review
python -m ipykernel install --user --name tgcm-review --display-name "TGCM reviewer"
jupyter lab

Large data and model assets: the full 822.44 GiB CAPTure CSV snapshot is not stored in this website repository. Its external download URL will be added separately. Additional model and extractor assets will be linked as they are released.