Can cleaning change both matchAll results and their indices?

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.

A repaired token adds a result and shifts another

The first fixture contains ID, a selected marker and seven, followed by a space and ID8. Before deletion, only the second token matches the fixed expression and its index is five. After deletion, the observer records ID7 with captured digit seven at zero and ID8 with digit eight at four. The plain reference already has both matches. A supplementary smile before ID7 yields index three, counting the smile as two UTF-16 units plus a space. The no-identifier control yields an empty list. Captures and full matched strings accompany every index so the new match is distinguishable from the shifted existing one.

This iterator answers a lexical matching question

We spread the matchAll iterator into an array using the global expression /ID(\d)/g, then retain each full match, first capture and index. These are invented labels rather than a real identifier specification. The earlier sentence-offset article uses linguistic segmentation; here the destination recognizes a fixed token shape and cleanup changes which tokens qualify. The stateful test article uses repeated test calls on one object; today we inspect all iterator results from one operation. We do not attach annotations, parse code identifiers or assume the expression captures every digit system. The cleaner has no knowledge of this token grammar and performs no index remapping.

Recompute matches against the edited text version

If a destination stores regex matches, associate them with the exact string and pattern that produced them. After a deliberate edit, compute the result set again and compare both membership and positions. Merely subtracting one from old indices would miss the newly valid ID7 in our first row. Code-point counts also cannot substitute for UTF-16 positions, as the smile control demonstrates. These downloads reproduce the specified four strings in the recorded runtime. They do not supply a general annotation migration algorithm, establish a production identifier schema or infer that a nonmatching token contains an unwanted marker. Review the intended token before removing its character.

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