Studying how hidden information breaks
Thesis
You can’t reason about the robustness of a watermark or a steganographic scheme until you’ve tried, carefully, to break it.
The steganalysis toolkit exists to answer one question empirically: how fragile is a given hidden-information scheme? LSB embedding, invisible Unicode, a visible watermark: each fails differently under pressure, and the only way to know how is to measure it.
The toolkit exists because I kept wanting the same measurement and rebuilding it ad hoc. The design constraint that mattered most turned out to be uniformity: one verification contract for every detector and cleaner, because without it two schemes can’t be compared honestly. The deliberate scope line (no defeating provenance standards like C2PA) is what keeps the work defensive.
Red-team work in service of blue-team understanding. The useful output isn’t a break; it’s a fragility measurement you can actually trust. That measurement posture rhymes with Publish Guardian and Sensitive Data Audit: prove the boundary holds, don’t narrate that it does. Scripts judge; agents propose is the general form.