About

The rebate engine is a stand-in

Every amount here was computed by a reference calculator written for this demo, not by Speedy’s rebate engine. It implements the two program kinds the spec names and nothing else: no tiers, no minimums, no accruals, no proration. Its name and version are stored on every row it produces and shown beside every number it computed, so swapping in a real engine relabels these screens by itself.

One rule produces an estimate, and says so

When a catch weight has been stripped from the export, the weight that was actually shipped is gone. It cannot be recovered, only estimated from the same item’s other lines. Those lines are badged “Estimated” in the drill-down and flagged in the export. That estimate’s error is the entire reason the explained figure reads 99.8% and not 100%.

The export is a proposed shape

The attributed-variance download is a documented, versioned JSON with a published schema — a plausible shape a downstream root-cause tool could be adapted to, not an interface anyone has agreed to. Saying otherwise would be the one dishonest claim in this project.

Two timings, and only one is a machine budget

“Upload to first report” is wall clock. On the seeded demo it has measured between 5 and 12 seconds, of which 3 to 7 seconds is this software working; the dashboard shows this instance’s own figures for its latest run. It shows both, because reporting only the first and calling it a pipeline budget would overstate what the machine does. Machine time excludes queue scheduling, which makes it the more meaningful of the two — but both move with how busy the machine is, roughly doubling on a host that is already working hard. Read them as orders of magnitude, not benchmarks.

The seeded demo has no person in it — reset approves the mapping from a committed spec, so nothing waits for a decision. A real onboarding’s wall clock is dominated by the decisions this one skips, and would be measured in minutes or hours.