New screenshots: one real session showing the family-naming model

Add the family-naming model as stage 4; remove the gpu/cpu runtime choice

show Loading under the report and lock the scan button while MiMo runs

example rail: buttons labelled by attack type down the left side

add 13 unlabelled example PDFs (12 families + clean) with developer key

fix stale batch dropdown when the slider moves; label regex hits as signatures

per-family treatment advice under payload verdicts; accuracy caveat in the intro

payload reports lead with a do-not-open / erase instruction

drop llama-cpp-python (unbuildable on Spaces); explain MiMo-over-Gemma choice

runtime picker (gpu/cpu); shorter GPU grant to stretch the free ZeroGPU quota

label marker vs sweep regions; explain sweep parse failures; sweep toggle

batch dropdown: accept values before choices are populated

batch the whole document into runnable chunks; user picks which batch to scan

ZeroGPU: duration under the 300s cap, weights prefetched outside the grant

dual runtime: @spaces.GPU 4-bit NF4 path (Part B config) + llama.cpp CPU fallback

MiMo-7B PDF injection detector: CPU/GGUF runtime, verbatim Part A+B logic