Make the next row easier to review.

The four-line sample flags the vendor URL containing bet and leaves three rows without a signal. Four equal one hundred seventy pixel blocks count four nonempty lines, with one highlighted. The flagged-row fraction is one of four, twenty five percent of this fixture, not a probability of harm or a verdict about a person or policy. This scan is a local substring heuristic, not workbook upload, automatic blocking or a gambling assessment. The source scans seven ordered terms but keeps just the first match in each row. A line with gamban, bet and casino therefore records gamban once, not three flags. Betting records bet because it appears earlier; alphabet also contains bet as a substring. Three equal term boxes and one highlighted match show this selection rule. These examples expose false positives and precedence without claiming activity was detected or blocked. In a hypothetical four-row scan with three matches, Flagged shows three and Clear one, while export writes all four. Common one hundred fifty pixels per row bars compare one visible with four exported. Review, Resolved and Escalate are manual dispositions; changing one saves it and does not remove its signal. Editing the text requires Scan to replace parsed records. Clear removes parsed rows, but after reload an empty saved text falls back to the sample while the saved empty review queue remains. Nothing updates an actual workbook or sends data.