Kirchenbauer partitioning uses key hash of previous token to split vocabulary into Green List (promoted by +δ) and Red List.
Simulate human editing, synonym replacement, or text paraphrasing to evaluate statistical detection persistence.
The invisible watermark workbench assigns an original logit from log of one plus an authored word hash modulo fifty. It adds delta only to green tokens. An illustrative green logit of one becomes three at delta two; a red token remains one. Both bars measure logits at one hundred sixty pixels per logit unit. These are synthetic annotations and the source does not generate fresh language model text. The written detector compares green count to token count times the assigned green fraction, then divides by square root of count times fraction times one minus fraction. With an illustrative one hundred evaluated tokens and fraction one quarter, expected count is twenty-five and standard deviation about four point three three. Thirty-five green tokens give z about two point three zero nine; forty-five give four point six one nine. Both bars depict z on one hundred thirty pixels per unit. These illustrative counts explain the algebra, not a measured false-positive rate or proof of authorship. The saved invisible lab includes the first word, using START token context. Ten words give ten evaluated tokens; twenty words give twenty. Both bars show counts at twenty-five pixels per token. A seeded pseudorandom generator fed by the preceding and current word hash assigns membership. Altering delta changes annotations but does not enter this membership rule. Its audit export preserves full inspected text and scores, without demonstrating actual cryptography, authorship or legal compliance.