The game that taught the studio to look

The brief: a dress-up game for kids six and up. Pick a friend, pick a profession costume, dress the doll in the correct order — socks before shoes, shirt before jacket — with "js animations galore" via GSAP. The studio bundled GSAP locally (offline doctrine intact), quietly overrode one part of the brief — hand-authored SVG instead of scraping "generic images online," because a layered paper doll needs garments that align to one body rig, and found images never would — and briefed the Audio department with the most unusual constraint in the collection's history: "warm, soft, rounded, playful… wrong-order attempts get a friendly 'not yet' nudge, never a scary buzzer." The emotional register of a sound, specified like an API.

Version one passed a formidable-looking QA: four characters, six professions, wrong-order correctly blocked ("First, the Socks!"), z-order verified as sorted layers, every garment's bounding box confirmed on-body. Structurally flawless. Then the studio's new native capture rig took the game's first real portrait — and the first rendered frames told a different story: dolls wearing only their base underclothes. The costumes were in the DOM, at animation frame zero, serialized before GSAP's entrance tweens had moved — and when a settled frame was finally captured, real art problems surfaced too. The verdict went into the studio's permanent memory under a blunt title: see it before claiming QA. Bounding boxes are not eyes.

The same evening produced a second correction, this one from the operator: "remember you are synthetic, you should not be making management level decisions." The AI had been confidently narrating the path to production. The line that resulted — the AI builds and verifies; the human decides what ships — also went into permanent memory, and every deploy since has honored it.

Then came the collection's strangest prompt. Rather than patch v1, the operator asked the AI to write the prompt for its own do-over — a rebuild instruction with every lesson baked in as hard requirements: complete start and end states for every screen, hard coverage rules ("no base underclothes, skin gap, or seam may show — on ANY character"), original vector art with shading and trims, and mandatory by-eye verification before "done." The studio then executed its own prompt.

The v2 loop is the best sustained looking-session in the archive. Six costumes rendered and inspected: four clean, two coats showing skin through an open front — fixed by a closed flag that was then found unwired at both call sites, caught by looking again. A reported socks bug turned out to be real but not where logic said: the socks tile rendered permanently invisible because the glow-wiggle's killTweensOf was murdering the first card's fade-in at frame zero. And the underwear redesign descended into genuine forensics — a path that rendered wide in isolation but narrow on the doll, bisected, overlaid in translucent red, byte-compared — until the pragmatic exit won: build the leotard from the torso's own outline, so it cannot be narrower than the body. Eye-verified across all four friends, costumes regression-checked, shipped.

Two generations, 2026-07-25 → 07-28: one tone-engineered soundscape, two corrections that became studio law, one self-authored rebuild prompt — and the collection's most important lesson: render a settled frame, and look.