Why does JSON.stringify omit the note attached to my cleaned-text array?

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.

Ownership does not guarantee appearance in an array export

The observer constructs a one-element array from the supplied text, then adds two enumerable named properties. Object.keys reports 0, note and 01, while length remains one. JSON serialization contains the element alone. The object-spread comparison includes all three keys in a JSON object, and hasNote confirms that the named metadata still exists on the original array. The omission therefore occurs during serialization rather than targeted text deletion. The leading-zero spelling is intentionally different from an ordinary array index, so it does not create a second indexed element.

An empty element still occupies its exported position

The empty-value control exports an array containing an empty string; the array does not disappear and length does not become zero. The numeric-text control retains a quoted 2, because the supplied element remains a string. In the marked fixture, the selected deletion changes interrupted AB to AB in both export views. The fixed note and leading-zero property remain present in memory at both stages. These controls separate element content, property ownership and export shape instead of treating every missing field as an invisible-character problem.

Give metadata a place in the transport schema

When a payload needs both text items and a note, design an object with explicit fields rather than attaching metadata to an array and expecting the JSON array syntax to carry it. Object spread is shown as a diagnostic comparison, not a recommendation to change an existing API from an array to an object. That change needs a destination schema decision. Our experiment uses no replacer, custom toJSON method, prototype modification or sparse element. It isolates the ordinary array serialization path for one indexed text item and two named properties.

Inspect the exact serialized document

The keys list is evidence about the in-memory value; the json field is evidence about the exported document. Keep both if you are diagnosing a lost annotation after cleanup. The cleaner receives only the text element and never creates the array or its metadata. Earlier collection coverage concerns Map and Set shapes; the new case is an ordinary Array with extra own properties. The authored note is not a real user annotation or a Claude artifact. Successful character cleanup cannot establish preservation of an application transport format, and these results do not measure model-level watermarks.

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