When is a robot-component specification safe to compare?
Compare a specification only after confirming the exact model or order code, the metric’s definition, the stated test or operating condition, the hardware boundary, the evidence state and the source date. If any material field differs or is unknown, keep the values separate and describe the gap instead of calculating a winner.
A specification can be accurately copied and still be unsuitable for comparison. “Maximum torque” may be a momentary endpoint while another supplier’s “rated torque” is thermal and continuous. A LiDAR range may refer to a high-reflectivity target under a stated light level while another number is only the device’s representable distance. Both numbers may be authentic manufacturer statements; they still answer different engineering questions.
RCI therefore treats comparison as an evidence-matching problem before it becomes a sorting problem. The workflow below is usable with an RCI record, a manufacturer datasheet, a distributor page or a test report.
Run six checks before putting values in one column
| Check | Question | Stop condition |
|---|---|---|
| 1 · Identity | Is the exact model, revision, lens, ratio, voltage or order code fixed? | A family name or marketing label covers multiple configurations. |
| 2 · Meaning | Is the metric continuous, rated, peak, stall, typical, maximum or representable? | The same unit hides different definitions or durations. |
| 3 · Condition | Are voltage, speed, temperature, target, illumination, return mode and duty stated? | A material condition is absent or differs between records. |
| 4 · Boundary | Does the value describe a bare motor, complete joint, sensor, module or integrated system? | The hardware included in the measurement is not equivalent. |
| 5 · Evidence | Is the value manufacturer claimed, independently verified, derived, conflicting or unknown? | A primary source is being mistaken for independent proof. |
| 6 · Time and conflict | When was the source checked, and does another credible source disagree? | The selected value silently replaces a newer, older or conflicting value. |
The stop condition does not make a component unusable. It changes the honest output. The result may be “not comparable from public evidence,” “comparable only under this condition,” or “requires a supplier clarification or bench test.” Those are decision-relevant findings.
Three current records show why the checks matter
Livox MID‑360: range needs reflectivity and illumination
The Livox MID‑360 record carries detection range as 40 m at 10% target reflectivity and 70 m at 80%, both under 100 klx ambient illumination. Copying only “70 m range” removes the condition that makes the number interpretable. All six current MID‑360 claims are M-state manufacturer statements from the same official product record. RCI preserves their conditions but has not independently measured the LiDAR.
Unitree GO‑M8010‑6: maximum torque is not continuous torque
The Unitree GO‑M8010‑6 record publishes a 23.7 N·m maximum-torque endpoint at a 6.33 reduction ratio with a 24°C measurement note. The cited page does not publish a continuous or thermally rated output-torque envelope, duty limit, cooling condition or allowable stall duration. The correct comparison output is therefore not “23.7 N·m continuous.” It is a maximum endpoint plus an explicit continuous-torque unknown.
maxon 734439: a conflict is data, not an editing inconvenience
The live 2026 catalog for maxon EC frameless DT 50 S ordering number 734439 lists 368 mN·m nominal torque, while a May 2025 catalog-page PDF for the same part lists 322 mN·m. RCI carries the live value and records the discrepancy as unresolved. It does not average the figures or silently discard the older source. The customer housing and heat path also remain material because this is a frameless motor kit, not a complete joint.
Use a comparison gate, not a completeness score
A row can enter a condition-matched comparison only when all six checks pass for the decision being made. Use this sequence:
- Define the decision. State the required operating point, environment, interface and hardware boundary.
- Normalize names and units without changing meaning. Unit conversion is allowed; converting peak into continuous performance is not.
- Match conditions field by field. Record every unmatched condition as unknown rather than assuming equivalence.
- Carry the evidence state into the output. A manufacturer comparison may still be useful, but it must remain labelled as manufacturer evidence.
- Escalate material gaps. Request a dated supplier answer, locate a qualified independent test or design a reproducible bench test.
Do not convert this gate into a generic percentage ranking. Missing fields are not proof of poor hardware, and a well-documented product is not automatically the best product. The gate measures whether the public record supports one bounded comparison.
Common questions
Does an official datasheet count as verified evidence?
It verifies that the manufacturer published the statement at the cited location. It does not independently verify that the product achieves the value in practice. In RCI, that normally remains evidence state M.
Can two values be compared if the units match?
Not by units alone. Definition, duration, voltage, speed, temperature, target, software mode and hardware boundary may still differ.
What should happen when a condition is missing?
Record it as unknown, explain why it matters and identify the smallest source request or test that would resolve it. Do not infer a convenient condition from a neighboring model.
Does an RCI component page prove compatibility or procurement readiness?
No. A claim ledger supports source tracing and bounded comparison. Compatibility, availability, certification and complete procurement readiness require additional evidence.
Evidence boundary, workflow disclosure and sources
This guide is EV2 editorial work: primary-source claims were transcribed into structured records and compared for identity, conditions and conflicts. RCI did not obtain hardware, reproduce the measurements or perform physical testing for these examples. Automated and AI-assisted tools supported retrieval, consistency checking and drafting; the cited records and upstream sources remain the evidence, and the RCI Editorial Team is accountable for publication.
Use the RCI methodology for M/V/D/C/U states and EV0–EV6 work levels, the current component catalog JSON for machine-readable claims, and the editorial transparency audit for the current verification boundary. Each example above links to its exact-model claim ledger and official source route.
Suggested citation: Robot Component Index. “Before You Compare Robot Components: An Evidence-First Verification Guide.” RCI 056, 2026-08-24. https://robotcomponentindex.com/research/before-you-compare-robot-components/