Marketplace Seller Grading System

Two blocks drive the grade: metrics, carrying onTimeDeliveryRate, customerSatisfaction, inventoryAccuracy, and policyCompliance, and sellerInfo with the account's id, name, and monthsActive. The four percentages and the star rating each feed their own scoring table in parallel.

Delivery performance is banded with a first hit policy on metrics.onTimeDeliveryRate: 98 or above earns 5 points, 95 earns 4, 90 earns 3, 85 earns 2, and the blank fallback row assigns 1 to everything below, writing the result to scores.delivery.

A high-90s on-time bar is realistic because large marketplaces hold third-party sellers to late-shipment thresholds in that range before restricting accounts. The single-point steps between 85 and 98 give sellers a visible ladder to climb, while anything under 85 is treated as operationally failing rather than merely weak.

Star ratings compress into five points through a first hit ladder on metrics.customerSatisfaction: 4.8 and above scores 5, 4.5 scores 4, 4.0 scores 3, 3.5 scores 2, and the fallback row gives 1, stored as scores.customerSatisfaction.

All the cutoffs sit between 3.5 and 4.8 because ratings on a five-star scale cluster high, so the meaningful differences between sellers live in the top of the range. A seller below 4.0 already reads as risky to shoppers browsing a marketplace, which is why the ladder punishes mid-range ratings harder than the raw numbers suggest.

Stock records are held to the strictest top band of the four tables: metrics.inventoryAccuracy needs 99 or above for 5 points, then 95 for 4, 90 for 3, 85 for 2, with the blank row assigning 1, saved to scores.inventory.

The tighter 99 threshold is deliberate, since inaccurate inventory turns directly into oversells and cancelled orders, the defect marketplaces penalize most heavily. The lower rungs mirror the delivery table so the two operational metrics read on the same scale when averaged.

Policy adherence uses the same five-point ladder as delivery: metrics.policyCompliance at 98 or above scores 5, then 95, 90, and 85 step down to 2, with everything else caught at 1 and written to scores.compliance.

Holding compliance to a 98 top band is a sensible business choice because violations, unlike a late parcel, expose the platform itself to listing takedowns and legal risk. Scoring it on the shared 1-to-5 scale keeps it weighted equally with the operational metrics in the average rather than acting as a separate veto.

Aggregation happens in one expression step scoped to the scores object via its outputPath: averageScore is avg(values(scores)), strengths collects metric names with filter(keys(scores), scores[#] >= 4), and improvements collects those at 2 or below. Producing named strength and improvement lists alongside the average is what lets the response tell a seller what to fix, not just how they rank.

Letter grades come off a first hit ladder over scores.averageScore: 4.5 and above is 'A' / 'Outstanding' and 'Eligible for premium seller status', 3.5 is 'B' with featured status, 2.5 is 'C' with 'Standard seller privileges', 1.5 is 'D' requiring a performance improvement plan, and the fallback row issues 'F' with 'Account review required'. With passThrough off, the response reduces to grade, status, and recommendedAction.

Half-point cutoffs around each whole grade mean a seller must average well on all four metrics to move up, since one weak score drags the mean below the next boundary. Tying grades to concrete privileges like premium and featured placement follows how marketplaces actually run seller tiering, and the D and F rows attach mandatory actions so poor grades trigger a process instead of just a label.