# Almost Human

> A playful uncanny-valley check. Name one almost-human thing - a doll, a robot greeter, a mannequin, an AI voice, a lifelike cake - rate how real each of its cues looks from 0 to 10, and see where it sits on the uncanny-valley curve and which cue to move first.

URL: https://almost-human.skillsafe.ai/
API: https://almost-human.skillsafe.ai/api.html

## What it does
- Free, in the browser: the uncanny-valley curve drawn as a wooden roller-coaster, a car parked at the thing's human-likeness, one flag per rated cue and a bracket for the spread (SVG and PNG downloads), plus the page's own placement and rule.
- Cues, in tie-break order: face, eyes, skin or surface, proportions, movement, voice, replies and timing. Human-likeness is the mean rating times 10; the spread is the highest rating minus the lowest. Under 60 is the cartoon side, 60 to 91 the valley, 92 and up the convincing side.
- Verdict, first match wins: match the cues (spread 5 or more); leave it be (outside the valley); finish the realism (in the valley, meant to pass as real); pull it toward cartoon (in the valley, any other goal).
- Metered task `valley-check` (gpt-terra): the cues that give it away, 1 to 3 concrete moves each marked less real, more real or keep, the cues already working and a short ride report. The page re-checks the verdict, zone, first cue, directions and numbers against its own.

## Input (all strings)
task, thing, where, goal, count, cues (one line per rated cue: `id: N of 10 | note: ...`), page_read (the page's numbers), optional retry_note.

## Notes
- The curve is illustrative, after the classic uncanny-valley sketch; ratings are the person's own estimates. A playful perception check, not professional design or psychology advice.
- Nothing is stored unless the person presses save; saved rides go to a per-account collection.

## Sources
- Skill: @dylantarre/appeal-mastery on SkillSafe - https://skillsafe.ai/skill/@dylantarre/appeal-mastery