No, K-beauty's Ulta shelf didn't jump six points in a week.

A number on this site moved six points in a week, and the honest thing to do is tell you it didn't. Our Ulta K-beauty reading went from 16.0% in W34 to 22.0% in W35. Nothing about Ulta's shelf changed that fast. What changed is how we read it.

The number moved because the scanner did, not the shelf

W34's Ulta chart came from a human screenshot run through OCR; W35's comes from a direct feed where every brand and product name arrives as text. The old way mis-read brand names badly enough that real products fell out of the count; the new way reads them cleanly. So the six-point "gain" is almost entirely brands that were always on the shelf but were invisible to last week's method — not new K-beauty arriving.

What last week's capture actually said

Here is the receipt. In the W34 saved capture, the #1 row read as "KOPAR KOPAB Jhelry Amco SURREAN BR" — an OCR smear of several stacked labels, not a brand. TATCHA came through as two separate brands, "TATCA" and "Tatcha." Rows like these were counted as not-K-beauty, and a few of them were the difference between 16% and 22%. The shelf didn't gain K-beauty; our count stopped losing it.

Why we block the comparison instead of reporting the jump

When the measurement method changes, what you are measuring changes with it — so the two weeks are not on the same axis. Rather than publish a jump we can't stand behind, we tag the channel with scope_changed_week: 2026-W35, and the code refuses to draw a week-over-week ranking or share comparison across that line. The W35 reading is the more accurate one; we just won't pretend it's a trend off a number we now know was wrong.

The rule: suspect your own pipeline first

A best-seller share that leaps six points in seven days is far more likely to be a measurement change than a market one. So our first suspect for any surprising number is us — our parser, our capture, our definitions — not the shelf. Most of the errors that would otherwise become a confident headline are caught right here, at the boundary between what a scanner read and what was actually on the shelf. That's the unglamorous work that makes the rest of the numbers worth trusting.