The actual clustering + basin-discovery-curve logic — real Python, executed in your browser on WebAssembly. Paste aperture reports (one per line); the Python decides coverage vs collusion vs unreadable from how distinct basins accumulate.
DEPENDENCY DISCLOSED: this loads the Pyodide CPython runtime (~6MB) from a CDN on first run. It is single-file but NOT offline — the first execution needs network. If the runtime can't fetch, this fails loud with a diagnostic; it never pretends to run.
Python runtime: not yet loaded (loads on first run).
0.55
— paste reports and run; the verdict is computed by real Python from the curve shape —
embedder: this in-browser build uses a token-hash embedder (offline, no model) — it catches verbatim/near echoes but NOT paraphrase. lower the threshold to merge more. for true semantic auditing use the CLI (swarm_audit.py --embed local) with sentence-transformers. the logic is identical to the CLI's: cosine-agglomerative clustering → cumulative distinct basins → curve-shape verdict. verified on CPython before shipping.