Independent measurement
A repeatable test or observation published by a party independent of the vendor and operator.
Evidence standard / Updated 2026-09-04
Evidence first.
Unknown stays unknown.
HumanoidUptime evaluates the quality and completeness of public reliability evidence. It does not infer robot performance from missing data.
01 / Evidence levels
The source determines the strength
Operator corroborated
A named deployment site or operator confirms the activity or result. Commercial involvement is disclosed.
Manufacturer reported
The robot maker publishes the value or description. It remains attributed to that maker.
Capability claim
A design intent or maximum capability without a defined field observation and denominator.
Not disclosed
No qualifying value was identified in the reviewed source set. This is not evidence of poor performance.
Not applicable is separate from evidence strength.
It means a field does not fit the bounded test or task in that record. A record-specific reason is required, and the same metric may still apply in another deployment.
02 / Definitions
Nine fields, defined
- Runtime hours
- Time the robot is publicly reported to have operated. Calendar deployment duration is not a substitute.
- Availability / uptime
- Available operating time divided by scheduled operating time, using a disclosed definition and observation window.
- Human interventions
- Human actions required to resume, complete or prevent failure, normalized by operating time or task volume.
- MTBF
- Mean operating time between explicitly defined failures.
- MTTR
- Mean elapsed time from a defined failure until recovery or return to service.
- Task output
- Completed units with task definition, attempt count, rejects and assisted cycles where available.
- Energy / charging
- Measured runtime between charges plus charging or swap downtime, preserving the operating scenario and test method.
- Autonomy mode
- Autonomous, supervised, teleoperated or mixed operation, with boundaries stated.
- Fleet size
- Number of robots represented by the reported evidence.
Availability normalization
A source label is not yet a comparable metric.
HumanoidUptime calculates technical availability only when the observation window, scheduled operating time, technical downtime and exclusion rules are public. Runtime, shift duration and task output are not substitutes.
(scheduled time − unscheduled technical downtime) / scheduled time03 / Scope and mission
A number belongs to a defined system
A result may describe an individual robot, a fleet, a task, a work cell, a deployment program, a product configuration or a production line. HumanoidUptime keeps that unit of analysis separate from the tested configuration and mission.
- identify the asset, fleet or wider system represented,
- preserve the supported hardware and software configuration,
- state the task, environment, object or payload and success rule, and
- show schedule, energy, control, human-support and safety boundaries as known, partial, unknown or not applicable.
Partially defined context is not completed through assumptions. Claims about another product version, site, stage or program remain adjacent context and do not become evidence for the bounded record.
04 / How evidence is assessed
Five checks before a claim appears
- 01
Trace the source
Identify the publisher, URL, publication date and source type.
- 02
Separate each claim
Assess each material statement on its own instead of treating an entire release as one fact.
- 03
Preserve the context
Keep the task, site, time window, fleet size, autonomy mode and limitations attached.
- 04
Show who supports it
Distinguish independent measurement, operator corroboration, manufacturer reporting and capability claims.
- 05
Keep it current
Show the review date and record material changes in the public evidence history.
05 / Comparison gate
When two numbers may be compared
Two models are compared on performance only when all five conditions are known and sufficiently aligned.
- the same metric definition and unit,
- a comparable task and environment,
- a stated observation window,
- a known fleet, attempt count or other denominator, and
- a comparable autonomy and intervention boundary.
Failure on any condition blocks the comparison. The disclosure matrix may still compare what is public, but not which robot performs better.
Review the current comparison gate →Corrections
Specific evidence wins.
Corrections are evaluated against the same source and context rules as every published record. A manufacturer may correct a factual error but cannot convert an attributed claim into independent evidence.
Submit a correction →Limits
Disclosure is not performance.
A sparse record may reflect confidentiality rather than weak reliability. A detailed record may reflect transparency rather than superior performance. HumanoidUptime keeps those ideas separate.