Why does a one-byte prefix decode as a replacement character after cleanup?

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.

Removing the marker does not make the next character one byte

The marked-accent fixture is encoded from the exact supplied Unicode string. Its first byte initially belongs to the three-byte U+200B. After deletion, the full UTF-8 buffer begins C3 A9 for é followed by the ASCII X byte. Taking just the first byte leaves C3, which Buffer converts to a replacement character. The plain éX control repeats this incomplete-prefix behavior with no edit. The emoji control begins a four-byte sequence and likewise fails when only its leading byte is selected. The ASCII control supplies a complete A byte and decodes successfully. We retain full buffer hex, selected prefix hex, decoded prefix, fatal validation and the complete first code point to locate the loss in the byte selection.

String boundaries and encoded byte boundaries are separate

Our earlier UTF-16 slice example cuts a JavaScript string unit, and the grapheme guide selects complete displayed clusters. This observer instead encodes the whole string before taking a Buffer subarray. A complete first code point may require more than one UTF-8 byte, as the accent and emoji controls demonstrate. The fatal decoder receives only that selected byte buffer; it is a diagnostic independent of the cleaner and does not show that the homepage validates uploaded bytes. The encoded buffers are constructed locally from valid strings, not recovered from a damaged file. We do not implement a general byte-budget truncator, test streaming chunk joining here or promise that a complete code point is always a complete visual character.

Apply a byte quota without discarding the remaining sequence

When a destination has a byte limit, calculate the encoded size and preserve boundaries required by its encoding and display contract. Do not decode an arbitrary fixed-size prefix and then treat the replacement character as recoverable original text: the omitted bytes belong to the sequence you cut away. Keep the complete encoded buffer or source when a later correction needs those bytes. Our four controls show why cleanup and byte truncation cannot be combined into a single success counter. The attached code uses local buffers and no filesystem or network write. It does not choose the correct quota for a service or certify all truncation algorithms. Test the actual final bytes and destination decoder after any deliberate prefix operation, with explicit error behavior and retained version metadata.

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