A used robot carries two independent risks and blending them into one number hides both. So we publish two.

The Condition Score /100 describes this specific machine, measured against the new-condition reference for its own model. It answers: how much of its original capability is left?

The Model Grade A–E describes the platform — parts availability, documented failure modes, vendor support horizon, reducer design. A well-kept arm on a dying platform is a good machine and a bad purchase, and only two scores can say that.

Certain findings are red flags that cap the Condition Score regardless of everything else: an identity mismatch, a crash logged without a subsequent recalibration, a joint temperature out of family under a known load. A cap is always printed with its reason.

Encoders sit on the motor side, before the gearbox. The controller therefore knows the motor angle with great precision, and knows nothing about where the arm actually ended up.

Wear happens after that measurement point: backlash in the strain wave or cycloidal reducer, lost preload in the bearings, flex in the structure. A robot can report itself perfectly on-target while the flange is a millimetre away from where it thinks it is.

This is the physical reason the Certified level exists. Repeatability is a property of the arm, so it has to be measured on the arm, with instruments that do not share the controller’s blind spot.

Repeatability is measured relative to a point P0 taught on the pendant, not to an absolute world frame — the question is whether the robot returns to the same place, which is what production actually depends on.

The robot then runs roughly 30 cycles that approach P0 along randomised trajectories, from different directions and speeds, and we record the dispersion of the returns. ISO 9283 is the base for the pose definitions and the statistics; NIST test methods inform the fixture and the approach patterns.

The arm is never loaded to 100% of nominal payload. A calibrated mass at 50–80% of rating loads the joints realistically without risking a machine that may already be worn — and a known mass is also what makes the current readings interpretable.

Every session is single-use and bound to the serial number of the machine under test. A session cannot be replayed, and a report cannot be moved onto a different arm.

The cycle intervals are drawn from a random sequence generated per session, which leaves a temporal fingerprint in the controller log. A fabricated log will not match it.

We read the controller’s native journals — written by the manufacturer’s firmware, not by us — rather than any file the seller can hand over. On top of that we check physical invariants: joint temperatures and currents have to be consistent with the declared mass and the observed motion. They rarely are, when something has been staged.

This is the odometer problem, and the industry has not solved it. The counter lives in the controller. Reloading the system software resets it to zero. Controllers get swapped between arms, sometimes for legitimate reasons.

So we never take an hour reading at face value. Every report carries a triple identity cross-check: the serial number the controller declares, the nameplate photographed and timestamped on site, and the firmware installation date. Three sources that a tampering has to line up simultaneously.

A swapped arm, a swapped controller or a reset counter shows up as a mismatch between them. It is the cheapest lie in this market, and nobody else checks it.

Every instrument in the kit carries a traceable calibration certificate, and the certificate reference appears on the reports it contributed to.

Each measured figure is published with its uncertainty. A repeatability of ±0.06 mm means nothing without knowing whether the instrument resolves 0.001 mm or 0.05 mm, so we state both.

Each report also states its own confidence level, which is capped when something could not be measured — a controller generation without live telemetry, an arm we could not run through the full cycle count, a cell that could not be cleared for the payload.

We do not predict failures. A condition assessment is a measurement of the present, not a forecast; anyone selling you a failure date for a reducer is guessing.

We do not certify absolute accuracy. We measure repeatability, which is what almost every industrial application actually depends on. Absolute accuracy requires a full kinematic calibration and is a different service.

We do not state a remaining service life in hours. And we do not state a market value — we describe condition, and the price is between buyer and seller.