metrics/*.jsonl: one row per sample (path, source, in/out bytes, kept%,
opaque in→out, and the input/output code), sharded under 10MB so the HF
Dataset Viewer renders a browsable in→out table without LFS. README card
points the viewer at metrics/*.jsonl and embeds the aggregate scoreboard.
Expand corpus to ~9.5k: 500 seeds x 19 obfuscation profiles + real samples
Replace the starter corpus with 500 deterministic seeds obfuscated under 19
javascript-obfuscator profiles (sharded by profile), plus 14 real scripts
crawled from top sites. Overall deobfuscation: 52.9% -> 4.6% opaque identifiers.
