# The Cyborg Index — methodology
> Answer summary
>
> The Cyborg Index scores humanoid robot offerings across 8 capability axes and 40 sub-axes, each anchored 0–5 against a written rubric and tied to a public source. Rank is set by Capability score. Sub-axes with no public evidence count as 0, so a robot cannot rank higher by publishing less.
> Current edition: August 2026 edition.
## The two scores
| Score | Definition | Formula |
| --- | --- | --- |
| Capability score | The Stage 4 cross-offering levelled score, 0–5, after rubric ceiling clamps. This is the score that ranks the Index. | levelled anchor, clamped by the evidence-kind ceiling |
| Proof score | Capability score scaled by evidence quality and breadth. A reference read on how well the public record backs the capability score — never ranked on. | Capability score × evidence quality × breadth |
| Proof gap | Capability score minus Proof score. How much of the score currently rests on claims and demos rather than independently verified material. Not a penalty and not a second score. | Capability score − Proof score |
| Evidence strength | Proof score as a percentage of Capability score. 100% means the public record fully backs the read. | Proof score ÷ Capability score × 100 |
| Coverage | How many of the 40 sub-axes carry a read from the public record. Low coverage does not soften the score — unevidenced sub-axes count as 0. | evidenced sub-axes ÷ 40 |
Rank is ordered by Capability score (desc). Proof score never re-ranks.
## Why unknown counts as zero
A robot cannot rank higher by publishing less. A sub-axis with no public evidence contributes 0 to the headline while still occupying its slot in the denominator. The cell itself is never shown as a scored zero — it renders as an em-dash in tables and a dashed tick on radials, because “we found nothing” is not the same claim as “it scored 0”.
Mean of 40 sub-axis scores, each 0–5. Sub-axes with no public evidence count as 0.
## The 8 axes and their sub-axes
### Locomotion
Flat, uneven, stairs, disturbance recovery, speed & endurance.
Sub-axes: flat_ground_walking, uneven_terrain, stairs, disturbance_recovery, speed_endurance.
### Manipulation
Grasp variety, bimanual, force control, in-hand reorient, tool use.
Sub-axes: grasp_variety, bimanual, force_control, in_hand_reorient, tool_use.
### Perception
Sensor stack, scene understanding, spatial memory, object permanence, world model.
Sub-axes: sensor_stack, scene_understanding, spatial_memory, object_permanence, semantic_world_model.
### Autonomy
Unattended operation, task horizon, generalization, instruction following, recovery.
Sub-axes: unattended_operation, task_horizon, generalization, instruction_following, recovery_from_failure.
### Learning
Imitation, sim-to-real, foundation policy, data flywheel, transfer & fine-tune.
Sub-axes: imitation_learning, sim_to_real, foundation_policy, data_flywheel, transfer_fine_tune.
### HRI
Speech I/O, NL instruction, intent recognition, safe proximity, social presentation.
Sub-axes: speech_io, nl_instruction, intent_recognition, safe_proximity, social_presentation.
### Deployment
Named pilots, field hours, customer diversity, supportability, commercial posture.
Sub-axes: named_pilots, field_hours, customer_diversity, supportability, commercial_posture.
### Safety
Functional safety, certifications, e-stop & failsafe, incident transparency, force/impact limits.
Sub-axes: functional_safety, certifications, e_stop_arch, incident_transparency, force_impact_limits.
## Evidence labels
| Label | Meaning |
| --- | --- |
| unknown | No public evidence found at read time. Counts as 0 in the headline. |
| vendor_claimed | Asserted by the vendor. Not independently verified. |
| reported | Reported by a third party, without direct observation by us. |
| demo_shown | Shown in a demonstration. A demo is not a deployment. |
| provisional | Read is supported but thin; expected to move as more material lands. |
| validated | Independently corroborated by material we can point at. |
| disputed | Credible sources disagree. Held open, not resolved by us. |
A vendor claim is never presented as an independently validated fact, and a demonstration is never presented as a deployment.

---

## Attribution

Source: Cyborgs of Earth — https://cyborgsofearth.com/methodology
Cite as: Cyborgs of Earth, The Cyborg Index methodology, 2026. https://cyborgsofearth.com/methodology
Framework: The Cyborg Index (8 capability axes · 40 sub-axes · anchored 0–5) — https://cyborgsofearth.com/methodology
Reuse terms: https://cyborgsofearth.com/licence. Free to quote and teach with attribution and a link.