Can direction controls remain after a no-findings result?

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. Invisible removal is enabled; the dash mode appears in each row. The observations below test this page's specific question. We make no detector-score or statistical-watermark removal claim.

What the no-findings message actually measured

The first three fixtures retain their complete strings and receive the same no-selected-character status. They contain U+200F, U+202E with U+202C, and U+2067 with U+2069 respectively. These controls are not in the five-character removal map. The fourth fixture contains both U+200F and U+200B; the cleaner deletes the latter, reports one removal, and leaves the direction mark. The data file records the status and remaining controls together so that a removal count cannot be mistaken for a complete inspection.

Stored order and display order are different

Unicode Annex #9 explains bidirectional display and the roles of directional marks, overrides and isolates. Our table prints escaped strings and code points in stored order, deliberately avoiding a display that could conceal the tested control. We have not rendered these fixtures in Arabic or Hebrew, inspected a malicious file, or measured accessibility behavior. Controls can be meaningful in mixed-direction text. Their presence alone neither identifies AI output nor justifies deleting them from a document.

Investigate the exact symptom

For an unexpected visual order, use the full Unicode inspector, compare stored code points, and identify the surrounding language and application. Preserve an original before editing a control, then check the display in the destination application. For source code or a filename that seems misleading, inspect the complete string rather than relying on its rendered appearance. The homepage performs no bidirectional-security audit and emits no warning for the retained controls in these fixtures. Its removal action has a precise limited scope. Do not interpret that status as a certificate that text is safe or free of all formatting controls.

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