How detection and text transformation work.

Text-watermark evidence is statistical, not visible markup

A generation-time text watermark can influence choices among plausible next tokens using a keyed, context-dependent rule. A compatible detector accumulates evidence over a passage; it does not locate a “watermarked word,” hidden Unicode string, or metadata field inside pasted prose. The general red-green-list construction and related schemes illustrate why a matching detector needs knowledge that is not visible in the text alone. Kirchenbauer et al.

Ordinary AI-authorship detection is different: it uses observable patterns in text to estimate similarity to a training distribution. It cannot recreate a provider's private key or production detector. Google's SynthID-Text publication describes a distinct watermark system, not a public specification for Claude. SynthID-Text publication

What the current score measures

The active local detector is claude-wm-local-v1. It creates a 0–100 signal from observable sequence features: character/token n-gram distributions, local token transitions, distribution variation, and consistency across overlapping text windows. The displayed score is an unweighted synthesis of those components and is labelled Signal score, never a provider probability.

Suitable English prose receives a score. Short, code-heavy, quote-heavy, unsupported-language, or otherwise low-evidence passages return Inconclusive rather than a misleading score.

Prepared, but not invented

Anthropic's transparency documentation says it has worked with industry and academia on watermarking developments and is preparing for applicable requirements. It does not provide this app with an official Claude watermark endpoint, request schema, credential type, or detection key. Anthropic transparency documentation

The code has separate detector adapters. The active local detector powers today's scoring. An inactive official-detector adapter is reserved for future documented integration; no placeholder endpoint or fabricated result is used.

Recompose from meaning, not from synonyms

Watermark removal uses one rewrite-provider request. The writer first forms an internal content blueprint covering claims, support, facts, names, values, links, terms, relationships, tone, audience, and intent. It then composes fresh prose from that blueprint rather than editing each source sentence in place.

The rewrite preserves the original tone automatically while protecting meaning, material facts, values, names, links, conditions, and technical detail. It is not guided by detector score and is never repeated until a score crosses a target.

Quality checks happen before the final comparison

Before a rewrite is returned, the server protects and restores numbers, dates, currencies, URLs, email addresses, code, and citations. It measures extractable factual values and named terms, phrase overlap, wording change, sentence/paragraph restructuring, and length change. These metrics are informational and do not reject a valid provider response because it is similar to the original.

When transformation succeeds, the application performs one independent local detector pass on the transformed text and displays before/after signals alongside rewrite quality, wording independence, structural change, and fact-preservation status. The comparison does not establish that an official provider watermark was removed.

Extensive editing can affect text-watermark detection systems generally; Google makes this point in its SynthID documentation. Google SynthID documentation

Focused processing, clear boundaries