Monorepo for Aesthetic.Computer aesthetic.computer

oskiewar: a second door on the title screen, and a wall behind it master

START is anywhere on the title — `tapTitle` makes the whole frame the button — which leaves nowhere to stand for a player who does not want to play the fighter but to be one. So the face is carved out of it. The game publishes the head's measured screen circle each frame and the shell hit-tests that ahead of the anywhere-button; clicking it opens the consent wall instead of starting a fight. The wall is the first slice of OSKIEWAR x REGARDE (vault: regarde/proposals/oskiewar-regarde). A player says in advance which parts of themselves the game may use, to make what, to be seen where, and for how long — before any material is collected and long before a model is invoked. Marketing, merchandise and model training sit in their own box, off, because they are not degrees of playing the game. Appearance does not imply voice, and the endpoint refuses an ask that assumes it does. REGARDE itself stays in the vault, under NDA. What lands here is a bridge: it shapes the answer into the DATA_OPERATION x GRANT_CONSENT request the desk expects and forwards it. Every branch that is not "allow" — unset gateway, timeout, unregistered term, deny — yields no capability, and the capability stays null until there is a worker whose reach a grant can bound. The Auth0 subject is hashed with a purpose-scoped salt before it leaves the process, so the desk learns a pseudonym and never the person. Two guards worth naming. The ring draws only when a wizard is actually mounted, so the Xbox and the reel renderer never show an affordance with no DOM to open; and only over a seat that is yours, because the attract loop's action variant dresses both fighters as bots and "add yourself" on a bot's head invites a player to replace somebody who was never there. The blackbox harness's file allowlist was already missing /account.mjs and /oskiewar-midi.mjs, which is why that suite fails at HEAD; completed it while adding the wizard, though the round-replay failure is older than this and survives the fix. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>