600me

AI-assisted response prepared for Kais (affiliated with Email Guard; see repository README for context). Thanks for the thoughtful feedback and for outlining those concrete acceptance cases. You've highlighted the exact architectural boundary: this experiment only tests a stateless signal policy and its HTTP response mapping (the 14 offline fixtures), rather than a stateful entitlement allocator. Handling timeout-then-success, duplicate hook events, and delayed results after cancellation—while ensuring "pending" never implicitly upgrades to "verified" and credits are allocated at most once—are crucial requirements for a robust trial orchestration system. Those scenarios sit downstream of the raw email signal check. We haven't built or deployed a persistent credit allocator or live hook handler in this showcase, so those tests aren't part of the current offline fixture suite, nor are they currently scheduled for this sample. However, documenting that separation between account creation and credit entitlement is valuable, and I've recorded these acceptance cases as design input for how signup signal consumers should structure their state machines.

One additional recovery boundary: keep bot-check failures separate from email-signal uncertainty. A transient DNS timeout should be indeterminate, not a reason to label the visitor fraudulent. Syntax, a disposable-domain list match, and DNS routing each answer different questions; none proves a mailbox exists. For a signup/trial flow, account creation and allocation of expensive credits can be separate server-side decisions: keep uncertain users limited, then apply eligibility and rate checks before granting costly usage. Authentication alone should not grant those credits. Direct Supabase Auth signups also need its native CAPTCHA enforcement; protecting a custom Edge Function does not cover that endpoint. Disclosure: AI-assisted contribution, affiliated with Kais Email Guard. This is an architecture suggestion, not a verified Supabase deployment or a claim about detection accuracy.