Elevator Problems

Build "Elevator Problems" — a polished, mobile-first portrait web app for the Eazo community.

CONCEPT
You are stuck in an elevator that stops at 20 floors, and each floor presents a two-option dilemma that gets more absurd as you descend. Your choices build a moral portrait.

RULES & STATES
- Two large choice buttons, always the same position, with the dilemma illustrated above by a code-drawn scene inside the elevator doors.
- Twenty dilemmas escalating from mundane (hold the door for a slow stranger, or not) to surreal (redirect the elevator into a wedding, or into a tax office).
- After each choice, a small counter reveals a fabricated but consistent "the previous 100 riders chose" statistic, computed from a fixed table so it never needs a network.
- Four hidden axes are tracked (mercy, chaos, pragmatism, spite); the descent visually darkens or brightens according to your running profile.
- Result: an elevator inspection certificate stamped with your moral archetype, your four axis bars and your single most damning choice. Share the certificate.

OPENING
No cover page: the app opens straight into the live interactive scene, fully playable on first paint. A floating title chip plus one animated gesture hint dissolve on the first touch anywhere on the stage. Keep gentle ambient motion (drift, breathing, particles, light) at all times so the scene is never a frozen image.

GLOBAL REQUIREMENTS
- Language: ship a lightweight built-in i18n layer with exactly two locales, English (en) and Chinese (zh). The app ALWAYS opens in English — never auto-detect, never read navigator.language, never default to a "system" preference. Provide one small, comfortably tappable EN / 中文 toggle at the top-right, about 56px down from the top edge, clear of the status bar and the dynamic island. Chinese appears only after the user actively switches. Share output stays English in both locales. Never mix two languages on one screen.
- Fully self-contained: absolutely no external APIs of any kind — no AI calls, no third-party data, media, image or font services, no remote backend, no login, no analytics. Embed every asset, dataset and piece of content directly in the app code (procedural drawing, inline SVG, emoji or code-generated art only).
- Real Share to the Eazo Community: on the result screen add a real Share button. Compose the result card as an image on a canvas and share it via the Eazo SDK community share — call share.compose({ text, attachments: [{ type: "image", url: dataUrl, caption }] }) and READ the resolved { accepted } value. Only when accepted is true may the button say "Sent to Eazo"; when accepted is false the button must read "Open in the Eazo app to share"; on a thrown error keep the original label. Never claim the post was shared on the plain web. Keep a Replay button beside Share.
- Layout contract (this is where most builds fail — follow it exactly): the app root must follow its host container height with position: fixed; inset: 0; height: 100% — NEVER 100dvh or 100svh, because the Eazo host injects a ~60px banner above the app and dvh overflows past the screen. Size every canvas from its PARENT element's clientWidth/clientHeight through a ResizeObserver, and run that fit ONCE on mount as well (a mounted component receives no resize event, so the buffer would otherwise stay at the 300x150 default). Convert every touch point through getBoundingClientRect() read at event time — never cache the rect, never use window.innerWidth/innerHeight for sizing or hit-testing.
- Safe areas: every top-side UI element sits at least 56px below the top edge. NOTHING you draw yourself — buttons, HUD text, score, hints, legends, progress bars, captions — may fall inside the bottom 88px of the viewport, because the Eazo mobile app renders its own native toolbar over that strip and your code cannot detect it. Push the play area up instead of padding it down.
- Fill the screen: the scene must occupy the full viewport height. No dead flat colour bands or empty background strips in the lower half — if the play area has a fixed aspect ratio, extend the artwork (gradients, ground, atmosphere, particles, reflections) into the leftover space. Reviewers reject apps whose bottom half is an empty block.
- No background music and no autoplay audio of any kind. You MAY add very short procedurally synthesized WebAudio blips (oscillator/noise, created lazily inside the first user gesture) as touch feedback, with a mute control; never load an audio file.
- Complete real states: loading, empty, success and error where relevant. Every control must actually work — no alert(), no console.log placeholders, no fake results, no button that does nothing.
- Content depth: give the experience enough material that it does not feel thin — aim for at least 8-12 distinct items / levels / variations of the core content, and make the difficulty ramp gently (the first 20 seconds must be winnable by anyone). Reviewers penalise "too fast", "too few", "only one interaction".
- Visual bar: exactly ONE core rule, understandable in 3 seconds, and generous visual response to every touch — motion, particles, deformation, colour, light, camera. Prefer canvas 2D; use Three.js with primitive/procedural geometry only if the concept is genuinely 3D. Do not use the canvas roundRect API (older iOS Safari lacks it). Never hardcode a light text colour that could land on a light background — derive text colour from the active theme.
- Determinism: never call Math.random(), Date.now() or locale formatting during SSR render — move them into useEffect or use a seeded generator, so hydration never mismatches. Finish with zero console errors at 393x852 portrait.