Method
How we measure, what we claim, and what we don’t.
What we read
Robots that follow the VDA 5050 standard publish their state, as it changes, to a message broker on the operator’s own network. We subscribe to it read-only. We install nothing on the robots and send nothing to them. Where the operator has them, we cross-check the robots’ reports against records the operator controls, such as the message broker’s connection log and warehouse task data.
What we reconstruct
From those messages we rebuild each robot’s time, minute by minute, into states that never overlap: productive, waiting for work, blocked, faulted, offline, charging, under manual control, scheduled idle, or unknown.
How we attribute
Every lost minute is assigned to the vendor, the facility, or no one, by fixed rules applied in a fixed order. We also look across the fleet. When robots from different vendors stall in the same place, the cause is usually the facility. When one vendor’s robots stall everywhere, it is usually the vendor.
When we don’t know
If the evidence does not support a conclusion, the minute stays unattributed, and we show that number as clearly as the others. A measurement that guesses is worth less than one that admits what it cannot see.
Whose definitions
Availability means different things in different contracts. We calculate it with the definitions in the contract between the operator and the vendor, not our own.
Evidence
Every message we receive is stored unchanged in a chain of hashes. Changing a single record breaks the chain, and the break can be checked by anyone holding the data. Every finding in a report links to the records behind it.
Reproducible by design
The engine is deterministic. No model guesses at the answer. Each report records the engine and rule versions that produced it, so the same data always gives the same result.
Neutral
We report what the facility caused as clearly as what the vendor caused. Independent means neutral, not against anyone.
What we never do
We never take money from a party we measure. We never control robots or act as a safety system. We never share one operator’s data with another; comparisons across sites use only anonymised, aggregated figures, and only with consent.
Limits
We can only measure what robots publish. Vendors implement VDA 5050 differently, and anything outside the standard is invisible to us, so every report states what it covered and what it did not. So far the engine has been tested against the official VDA 5050 schemas, on simulated fleets, and on 72 hours of real operations from one robot in a research laboratory. It has not yet run on a customer fleet.