Checked service, open-source tool you run yourself

guillaumemeyer/watermarks-remover, rewrite layer review: the mark was not found, 5 September 2026

The marked sample went through guillaumemeyer/watermarks-remover, rewrite layer on 5 September 2026, and the mark was not found in what came back: -1.1. 1 of 3 facts kept, 64% of the length, 79% of the meaning.

Address: https://github.com/guillaumemeyer/watermarks-remover

What came back, in five figures

Sample UM-D150B2, 992 words, 2026-09-05

The mark

-1.1

Detector score, our key. Mark not found: found at 4.0 and above, grey band down to 2.0.

Meaning kept

79%

Sample and result cut into pieces and matched piece by piece, so moving paragraphs around costs nothing.

Facts kept

1 of 3

Numbers, dates, links, names and terms taken out of the sample and looked for in what came back.

What did not come back

2, WW2

Wording changed

77%

Share of the sample's words outside the blocks that stayed identical. Longest untouched run: 6 words.

Length

64%

Words back against words out. The halfway tick is the length of the sample itself.

Layer A plus layer B: rewrite_text.py against an OpenAI-compatible endpoint on open weights, default tactic "paraphrase", one candidate, one loop.

Our own processing on the same sample, 5 September 2026: mark not found at 0.8, 3 of 3 facts kept, 92% of the length, 74% of the meaning.

Its row among the Claude watermark removers .

Test guillaumemeyer/watermarks-remover, rewrite layer yourself. The same sample, our detector on what comes back. Take the marked sample