Why can encodeInto write zero bytes for a cleaned emoji?

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 destination can fill before the source is consumed

The first synthetic string begins with U+200B followed by éX. Its three-byte marker fills the entire destination before cleanup: read is one UTF-16 unit, written is three bytes, and the remainder is éX. After the actual cleaner removes that marker, the accented letter and X together fit exactly. Their read count is two while their written count remains three. The plain accent control confirms that result without deletion. The emoji control cannot fit its first character, so neither count advances. Its zero-filled buffer is allocation state, not encoded text. The ASCII control consumes all three units and writes three bytes. We save both the whole buffer and the actually written prefix to expose that distinction.

Capacity reporting differs from cutting an existing buffer

The earlier byte-prefix experiment encodes first and then deliberately slices an incomplete character. Here TextEncoder receives a limited destination up front and stops without writing a partial leading emoji. These are different operations even though both involve byte limits. The remaining string is selected with the returned read count, whose unit is UTF-16 rather than bytes. We use a fresh three-byte Uint8Array for each observation. No network backpressure, reused buffer contents, streaming transport or retry loop is tested. The recorded empty written prefix must not be replaced by the full allocation when deciding what data was produced. Our observer does not change the cleaner or claim that its homepage exposes an encodeInto interface.

Track progress before retrying a remaining source

A caller that keeps offering the same undersized buffer to this emoji would make no progress under the measured conditions. Select capacity and retry behavior according to the actual encoder contract, and send only the written region onward. Retain any unconsumed source rather than treating a returned object as completion. The accent example also prevents confusing the read count with a byte offset: using two as the written length would omit the final X. These four local cases do not establish a universal buffer allocation strategy or a safe display truncator. They provide concrete counters to check in a destination integration. The downloaded harness preserves exact byte hex, remaining text and runtime metadata for independent reruns.

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