Why does my cleaned é decode as i through Buffer ascii?

Measured inputs and outputs

These are locally constructed test strings, not evidence that Claude inserts these characters. We executed the saved homepage script snapshot with a minimal DOM harness and compared exact strings. The invisible-checkbox state and dash mode appear in each row. The observations below test this page's specific question. Destination observations run independently on the exact input and cleaned strings in the recorded Node runtime. We make no detector-score or statistical-watermark removal claim.

The retained accent reaches two different decoders

The observer first generates Latin1 bytes from the exact string, then decodes the same bytes under two names. After cleanup, é is still the string sent to Buffer.from and its byte is E9. The Latin1 decoder returns é while the ASCII decoder returns i, recorded as code point 105. Thus the visible accent loss takes place after the cleaner, at decoding. Before cleanup the extra selected marker contributes byte 0B in this deliberately lossy encoding path; that byte disappears from the cleaned branch while the E9 disagreement remains.

Hex and decimal code points expose invisible output

Plain é reproduces the mismatch without a selected character. The ÿ control encodes as FF and ASCII decoding produces code point 127; its displayed glyph is not a reliable way to identify that result, so the record includes decimal code points. AB supplies an ASCII-only reference with bytes 4142 and identical decodes. Earlier Latin1 coverage measures loss when encoding a Unicode string; here the retained E9 and FF bytes already exist and the two decoders interpret them differently. No file, HTTP header or browser charset is consulted in this test.

Preserve bytes and label the conversion stage

When investigating an export discrepancy, keep the original Unicode value, generated hex and exact decoder name together. A cleaner status describing selected characters cannot certify that a later encoding roundtrip preserves every retained letter. The supplied reproducer uses Node Buffer behavior, not a browser TextDecoder with a similarly named label. It tests just four strings and does not recover information already discarded by encoding or high-bit clearing. These strings were authored for diagnosis. They do not show that Claude emitted any byte sequence, identify an author or measure removal of a statistical watermark.

Reproduce this test

Save reproduce.cjs and tested-app.js in the same folder. Run the command below with Node.js. The harness prints its runtime, script SHA-256 and every measured row. Compare those rows with the original record. Using a newer script or runtime creates a new experiment; retain the version information with your rerun.

node reproduce.cjs