★★★★☆
“Listing to Loyalty’s PDP-to-payment week exposed variant confusion we had labelled as ‘price sensitivity.’ The workbook format is fussy. The finding was not.”
What people said after the leak ranking landed.
These notes refer to specific Mirror Cloudcore modules. We keep one reservation in the mix on purpose — a studio that only collects praise is not a serious room.
“Search Ranking Diagnostics gave our discovery pod a tree they still keep on the war-room wall. Zero-result handling was the unglamorous win.”
“Cohort Retention Mapping refused to let us smooth payday spikes. Annoying in the session. Correct for a Thai wallet-heavy marketplace.”
“The seller-economics module assumed we already had clean take-rate data. We did not, so week two of Seller Take-Rate Lab dragged. After we rebuilt those extracts, the buyer-trust charts finally matched what support was hearing.”
Case note · grocery app, Bangkok
A grocery marketplace came in celebrating record GMV during a mid-year campaign. Module 05 asked them to split first-time buyers from second-order windows. Campaign GMV was up; the 28-day repeat rate among new buyers had slipped by a visible margin.
The leak ranking they left with was unromantic: stop stacking free-delivery codes on first baskets, instrument “promise vs. actual slot,” and treat seller stock-outs as a conversion metric, not only an ops KPI. Six weeks after the studio they reported the repeat window stabilising. They did not claim a miracle — only that arguments in planning meetings now used the same dictionary.
Case note · fashion resale, remote seat
A resale app’s search conversion looked “fine” until the diagnostics tree separated branded queries from long-tail. Long-tail was dumping people onto thin listings. Ranking changes without inventory quality rules made it worse.
In the experiment studio they fenced a listing-quality badge test so sellers outside the treatment could not instantly copy the badge. The point was not a huge lift. It was learning that a two-sided test without a fence just teaches the control group to mimic the treatment.