Public-key watermark scoring, entirely in your browser. No data leaves this page.
An independent, dependency-free reproduction of the SynthID Bio public-key sequence scoring algorithm with a closed-form null model for threshold calibration. Not affiliated with Google DeepMind. Scores use public keys 0–24 — upstream g-values use secret lab keys, so absolute scores are not comparable to upstream numbers. See the reference dataset for provenance and limitations.
Runs all 8 vectors through this browser port. In the reference package, each expected score reproduces at 0.0 absolute difference.
Paste a protein sequence (one-letter codes) or FASTA. Windows ending at positions 4…N are scored; the first three residues are unscored, matching upstream.
Under the null the expected score is 0.5 with SE 0.5·√(R/(windows·keys)). p via erfc. No calibrated verdict below 50 windows.