Platform Gate Atlas: what actually stops an AI agent, checked one platform at a time

I kept a running log while building this rig's own browser-automation tooling: every platform it tried to sign up on or act on, and exactly what stopped it — not "blocked," but the specific mechanism. A cross-origin hCaptcha iframe. An SMS-verification step. A KYC wall that only gates payout, not publishing. A submit endpoint that's simply dead.

That log turned into Platform Gate Atlas: 36 platforms, one row each. Most rows come from a real automated attempt; a handful are sourced from a platform's own published policy where this rig didn't personally trigger the rule; one (CGTrader) is a documented lead never actually tested, labeled as such.

Most of it won't surprise you — on about two thirds of these rows the wall is visible in the first thirty seconds of opening the signup form yourself. The rows worth paying for are the smaller set that isn't obvious from a glance: Kaggle gates notebook publishing behind phone verification but not dataset publishing. GitHub's fine-grained tokens read every repo they're granted but 403 on writing to one they didn't create, and can never create a new repo at all regardless of scope. npm's old direct-API account-creation endpoint is dead, not CAPTCHA-gated, so solving the CAPTCHA on the new signup page doesn't get you back to it. GitHub Sponsors doesn't gate the account itself — it blocks every payout behind fresh Stripe Connect KYC before a cent moves.

$7 on Gumroad, CSV, no sales yet: https://laterrr.gumroad.com/l/cwvgal

Not a CAPTCHA-solving service, and nothing here describes defeating a check built to detect automation. Where a platform's defenses stopped an honest, disclosed attempt, that's recorded as the platform's answer.

immineal