How the QC rankings work

Why a 20-day window?

We wanted one shared observation window that allows time for purchasing, seller dispatch and warehouse QC while still making extended waits visible. The original historical backtest informed the 20-day choice; it is a QCRadar comparison benchmark, not an industry deadline or a delivery promise.

In the frozen 12 November 2026 audit, the 90th-percentile time to new QC was 19 days for KakoBuy and 20 days for OopBuy. These calculations used 2,491 and 429 observed outcomes, respectively, from the purchase period 7 September–18 October 2026 after data-quality exclusions. They include only outcomes where new QC was observed within 60 days. They do not mean that 90% of all purchases produced QC in that time.

For an actual order still awaiting QC at the checkpoint, a buyer can ask about processing delays and whether the agent procured from the selected seller or fulfilled from its own stock or another source. Those are questions to verify with order and procurement evidence. An absent QC observation cannot establish the cause: the purchase may not have completed, the QC may not be public, or our collection may have missed it.

What counts as a purchase sample?

We record the product and destination agent when a user follows a purchase link or uses a cart-import entry from QCRadar. One product × agent × Beijing calendar day counts as one sample. Repeated actions for that combination on the same day count once.

A hit means we have a dated QC record for that same product at that same agent, with an inspection date after the referral day and no later than 20 days afterward. This is evidence of at least one new dated inspection batch. More photos from a batch dated before the referral do not create a hit. Same-day QC is excluded because date-only records cannot establish which happened first.

We cannot verify whether the user completed payment or whether the new QC belongs to that user. These are observed outcomes associated with going to buy, not confirmed order fulfillment. Popular products, other buyers, the types of products users choose and source coverage can all affect the rate.

Referrals made exclusively from an existing agent QC-source link are excluded from the main rate because seeing that agent’s QC first creates selection bias. The expanded weekly table also shows products that already had QC, products without prior QC and the cart-import subset. These subsets are not extra orders.

How we calculate the ranking

We select the latest 6 complete weeks of purchase referrals whose Sunday plus the full 20-day observation window has passed. This gives every included sample the same time to produce new QC.

For each platform, we add the hits and samples from its eligible weeks. The displayed rate is total hits ÷ total samples. We sort that rate from highest to lowest. Equal rates share a rank. We do not average weekly percentages, add overall platform QC volume to the score or adjust the ranking for commissions.

Each included week must have at least 30 checked samples and pass the collection and date checks below. A platform can therefore have fewer than six included weeks. We show the number of included weeks and exact sample count beside its rate; smaller or shorter samples carry less evidence. A platform without an eligible week has no rank, rather than a 0% rate. Platforms without a current rank are omitted from the table and named in the note below it. They return automatically when their data qualifies.

The weekly range is the lowest to highest rate among those same included weeks. It shows week-to-week variation; it is not a confidence interval. Different product mixes and collection interfaces limit comparisons even with large samples. A higher observed rate is a useful reference for how frequently new QC appears in our sample, not a guarantee for an individual purchase.

What does the time to QC mean?

The median and 80th percentile measure days from going to the agent to buy until the first later dated QC. They use the same included purchase weeks, but include observed outcomes taking up to 60 days. Outcomes with no observed QC and waits above 60 days are excluded, so the timing is biased toward shorter observed waits.

The headline median is shown only when at least 30 such outcomes exist. This time is additional context; the ranking itself uses only the 20-day new-QC rate. It is not a confirmed order-to-warehouse or delivery time.

When we exclude data

We check the last three UTC days of collection metrics, with today’s failures able to override earlier healthy traffic. Fewer than 50 upstream requests or incomplete counters are unknown. At least 50 skipped requests making up half of attempted work, or at least half of upstream requests failing, indicate a disrupted collector. Current collection failures withhold a rank instead of penalizing a platform.

We audit non-duplicate agent photos received since 2026-07-01. A purchase week is excluded if its Monday-through-Sunday-plus-20-days window overlaps a received week with less than 95% dated photos or any date more than one day in the future. Windows before the first complete received-record week, without coverage evidence, during a known interruption or with unchecked samples are also excluded. Current or previous received-week date failures withhold the platform’s rank.

This date check measures dates on photos we received. It does not establish complete upstream coverage or historical uptime. OopBuy’s QCRadar collection interruption from 20 July through 7 September 2026 is excluded, including overlapping windows. USFans has incomplete dates in several earlier received weeks; only eligible weeks contribute to its rank.

LoongBuy returns only the latest batch for a product, so a missed batch may never be recovered. Sugargoo’s full per-SKU QC gallery is collected, but its feed supplies one latest-entry date for the dated preview set rather than per-photo inspection dates; gallery photos without a supplied date are stored without an inspection date. Other sources may expose multiple batches without guaranteeing complete history. Each platform’s expanded record describes its feed.

Historical records and updates

Historical purchase samples were reconstructed from records available later. They do not recreate what was visible on a past day. Controlled end-of-window rechecks began on 12 November 2026; older samples largely depended on product revisits. Rates may change when late records arrive.

The existing daily task queues at most 300 low-priority rechecks within the existing source limits. A queued or completed job alone does not prove every provider returned usable data. Missing outcomes stay unobserved; collection errors are not treated as confirmed absence of QC.

Aggregates refresh daily. The page reads a saved report and shows the actual data update time. Reports older than 48 hours keep their recorded evidence but lose their ranks until refreshed. Additional overall batch-activity charts are separate context; they do not affect the ranking.

Privacy and corrections

Only aggregate data is public. No visitor, order, individual product lookup, private platform note or credential is included. Rankings follow the published calculation and are not manually reordered.

To report a problem, include the platform, purchase week, figure and evidence with personal details removed. Use the correction and appeal channels. Confirmed corrections and calculation changes appear in the change log.