Methodology
How we score
Every machine in the directory carries a Realism Index — our editorial estimate, 0 to 100, of how close it comes to a lifelike human companion. Here is exactly how that number is built.
The index is an editorial judgment, not a lab measurement. Nobody can put every machine on this site in one room — some ship today, some are prototypes on the far side of the world, some are still renders. What we can do is hold them all to one consistent rubric, score only what the public evidence supports, and show our working. This page is that working.
Six dimensions, weighted to 100
Each companion is scored 0–100 on six dimensions. The index is the weighted average, rounded to the nearest point. The weights encode an opinion we hold openly: for a companion, the face and the body matter more than raw athleticism — presence is the product.
| Dimension | Weight | What we're looking at |
|---|---|---|
| Face | 25% | Facial realism and expression — skin, eyes, gaze, micro-movement. |
| Body & skin | 20% | How lifelike the body reads — materials, proportions, touch. |
| Motion | 20% | Locomotion and body movement — gait, balance, fluidity. |
| Voice | 10% | Speech quality, lip sync and conversational latency. |
| Mind | 15% | Autonomy, memory and how convincing the intelligence is. |
| Reality | 10% | How real the product is — shipping truth, demos vs. deliveries. |
Face · 25%
Facial realism and expression — skin, eyes, gaze, micro-movement.
- Scores low
- No face at all, or a status display where a face would be — most industrial humanoids score near zero here.
- Scores high
- Silicone or Frubber craft with fluid micro-expression, believable eyes and gaze that tracks you — sustained, not just in a press photo.
Body & skin · 20%
How lifelike the body reads — materials, proportions, touch.
- Scores low
- Exposed chassis, industrial shells, or a body that exists only from the waist up.
- Scores high
- Lifelike skin over human proportions, warm or soft to the touch, convincing at conversational distance.
Motion · 20%
Locomotion and body movement — gait, balance, fluidity.
- Scores low
- Static or wheeled; a machine that must be carried to its spot.
- Scores high
- Confident bipedal walking, balance, and movement fluid enough that you stop noticing it.
Voice · 10%
Speech quality, lip sync and conversational latency.
- Scores low
- Canned lines, obvious latency, lips that don't match the words.
- Scores high
- Natural speech with tight lip-sync and response times short enough to feel like conversation rather than a query.
Mind · 15%
Autonomy, memory and how convincing the intelligence is.
- Scores low
- Scripted demos, teleoperation presented as autonomy, or no interaction model at all.
- Scores high
- Genuine on-board conversation with memory of you, running unscripted in other people's hands — not just the maker's.
Reality · 10%
How real the product is — shipping truth, demos vs. deliveries.
- Scores low
- A render, a concept, or a stage prototype with no path to your door.
- Scores high
- Shipping in volume to real customers, with third-party and owner evidence that matches the marketing.
The evidence rules
- Public evidence only. Scores are built from official demos, spec sheets, third-party footage, hands-on reporting and owner accounts — sources are listed on every profile.
- Claims are not capabilities. A manufacturer figure nobody outside the company has verified is treated as a claim, flagged as one in the profile, and discounted in the score until it survives contact with independent evidence.
- Choreography is discounted.Stage demos are rehearsed and teleoperation is routinely presented as autonomy. Unscripted footage in other people's hands outweighs any launch video.
- Reality is a dimension. A concept can describe a perfect machine; until it ships, its Reality score keeps the index honest. This is why a shipping industrial robot can outscore a beautiful render.
When scores change
Scores are revisited when the evidence changes — a major demo, a shipping milestone, credible owner reports, a delay. Each profile shows when it was last updated, and where we've recorded a machine's dated promises against what actually happened, you'll find a paper trailsection on its page. We don't retro-edit history; we add to it.
Independence
No manufacturer pays for a score, sees one early, or edits one. gfs.ai takes no advertising or sponsorship from the companies whose machines it scores; where an affiliate link ever appears, it's disclosed and has no bearing on the index. Read the full disclosure.
See the index in action across the directory, put any two machines side by side, or start with how realistic these machines actually are.
The briefing
The future of companionship, in your inbox
One careful dispatch a month: every price change, ship-date slip and new machine we log — plus the milestones that actually matter. No hype, no spam.
Free. Unsubscribe anytime.