# Stick the Landing

> A playful judging panel for one small everyday feat - carrying the shopping in one trip, a tray of mugs down the hall - written as moves, one per line, with the last move as the landing. One metered run gives every move its demands and a difficulty value, marks the faults with the person's own words, calls the landing (stuck, step, hop or fall), gives a remark from each of five execution judges and one tidy-up for next time. The page does every sum.

URL: https://stick-the-landing.skillsafe.ai/
API: https://stick-the-landing.skillsafe.ai/api.html

## What runs where

- Free, in the browser: a balance beam with one chip per move (its value letter A to E, a flag for any fault, an arc for each connection bonus) and a landing mat, an arena scoreboard (D + E = total), five judges' flip scorecards with the highest and lowest struck out, judging it yourself (a value, a fault size and lens per move, and the landing), the page's free read of demand and mishap words, and SVG, PNG, Markdown, CSV and JSON downloads, a printable card and a Share button (the score line and the card through the device's share sheet, where the browser has one).
- The loop, free: call the panel again on a rewritten routine (or on a saved one you reopened) and the scorecard says what changed since the last call - the total, D and E, the landing, and which marked faults went away or appeared.
- Metered, one task (`routine`): demands per move from five ids (carry, footing, both, clock, aim) and the value from their count; faults on moves before the landing, sized fall, large, medium or small, each with a judge's lens and an exact quote; the landing call with its quote and reason; five remarks, seats E1 to E5 (form, rhythm, precision, composure, commitment); a tidy-up and a sign-off. Runs on gpt-terra.
- The scoring: D is the sum of the values (A 0.1 ... E 0.5) plus 0.1 for each neighbouring pair both valued C or more with no fault on either. Each judge takes off 0.1, 0.3, 0.5 or 1.0 per fault, 0.1 more on a fault in their own lens, plus the landing (0, 0.1, 0.3 or 1.0); the highest and lowest are dropped and E is 10 minus the mean of the middle three.
- The page re-checks every reply: one element per move, each value against its demand count, faults only before the landing and quoted from their own move, the landing quote, each judge's move and the tidy-up move by fixed rules, word limits, and no score talk in the prose.

## Contract

- Input (all strings): task (`routine`), tone (`deadpan`, `warm`, `stern`), title (or ""), moves (`n | move` lines, 3 to 8, the last is the landing), page_facts (moves, landing_line and word hints), optional clip_note and retry_note.
- Reply: one JSON object with status (`ok` or `declined`), reason, title, elements (line, demands, value, why), faults (line, size, lens, quote, note), landing, landing_quote, landing_why, judges (seat, line, remark), tidy (line, tip), closing.
- Declined: routines that name a real person, are risky or unlawful (riding or driving while doing something else, heights, fire, knives, alcohol, drugs), involve a baby, a child or an animal, or are sexual, hateful or about self-harm.

## Sources

- Built on the SkillSafe skill @neolabhq/reflexion-3acritique: https://skillsafe.ai/skill/@neolabhq/reflexion-3acritique (specialised judges each review through their own lens, then a consensus settles it), applied to five execution judges over the person's own moves with the page dropping the high and the low.
- Saved routines live in the person's own SkillSafe account only when they press save.

For fun only - not coaching or professional advice. Never attempt anything risky for a score.