Your courier said "weather delay." How we'd check whether it rained.

A Non-Delivery Report is supposed to mean the courier tried and failed: customer unavailable, address not found, weather delay. The merchant pays for the failed attempt either way — and almost nobody checks whether the attempt actually happened.

Before anything else, the honest part: ShipSense has no paying pilot yet and no real merchant shipment book. We have not audited a customer, recovered a rupee, or won a dispute. Everything below describes what the engine tests for and how it decides — not results we have delivered for someone. When we have real recovered freight to report, we will report it with the merchant's name attached and a randomised holdout behind it.

With that said — here is what a failure report has to survive.

1. Weather claims against the weather

"Weather delay" is a testable claim. It names a place, a date and a condition. Rainfall for a given pincode on a given day is a matter of public record, so a weather-attributed failure either matches the record or it does not. A courier logging a monsoon delay on a dry day has written something checkable, and wrong.

2. Timestamp physics

Multiple "customer unavailable" attempts logged at the identical second cannot all be true. A field agent cannot ring twelve doorbells simultaneously. Clustered identical timestamps across different shipments are the cheapest fake-attempt signal there is, because faking a delivery attempt is a data-entry action and data entry leaves a rhythm.

3. Reason flip-flops

The same shipment cycling through "address not found" → "customer refused" → "address not found" across attempts contradicts itself. If the address was found on attempt two, it did not vanish by attempt three. The contradiction is the evidence.

What the engine actually knows so far

One thing we have measured, on our own test data rather than a customer's: the risk of an order ending in RTO does not climb smoothly with failed attempts. It jumps. Between the first failed attempt and the second, the probability of the order coming back rises several-fold. That is where an intervention is worth making — and it means chasing an order after the third attempt is mostly theatre.

That finding comes from synthetic data and will be re-run the day a real book exists. If it does not hold, we will say so.

Why this matters more than the percentage

Every merchant knows their RTO rate. The question nobody asks is how much of it is real. A failure the courier invented is not a cost of doing business — it is a billing error with evidence attached, and it is the only part of the RTO number that is arguable.

Want these checks run on your own shipments? The free 7-day audit does exactly this — CSV in, evidence out, and a straight answer about what we can and cannot see in your data.