Subject: Authentication implementation repeatedly reported as successful, but runtime behaviour is incorrect
Hello Google AI Studio Support Team,
I would like to report an issue regarding AI Studio’s implementation and debugging of Supabase Authentication in my Expo React Native application.
Over the last several days, I have spent many hours asking AI Studio to implement and repair a standard Supabase email/password authentication flow. Although AI Studio repeatedly reported STATUS: PASS and stated that authentication was fully working, my real-world testing consistently showed otherwise.
Project Information
- Framework: Expo React Native
- Backend: Supabase
- Authentication: Email & Password
- Environment variables:
- EXPO_PUBLIC_SUPABASE_URL
- EXPO_PUBLIC_SUPABASE_ANON_KEY
Both environment variables are correctly configured and loaded.
Problem Summary
AI Studio repeatedly claimed that:
- Sign Up was working.
- Sign In was working.
- OTP verification was implemented.
- Session persistence was working.
- Navigation was working.
- Runtime tests passed.
However, the actual application behaviour was completely different.
What Actually Happened
When testing the application:
- Creating an account produced little or no visible feedback.
- The Verify Email screen did not behave as expected.
- Sign In did not work correctly.
- The application often remained on the authentication screen.
- In several cases AI Studio reported PASS even though the feature was clearly not functioning.
The generated implementation reports did not match the application’s real runtime behaviour.
Investigation Performed
To determine whether the issue was with my Supabase project or with the generated code, I manually investigated every layer.
I verified:
- Supabase project URL
- Supabase anon key
- Environment variables
- Email provider configuration
- Authentication settings
- Auth logs
- SQL queries
- auth.users
- auth.identities
I also repeatedly rebuilt and restarted the AI Studio preview.
Confirmed Findings
The following has now been confirmed.
Environment
- EXPO_PUBLIC_SUPABASE_URL is present.
- EXPO_PUBLIC_SUPABASE_ANON_KEY is present.
- Runtime validation succeeds.
Supabase Connection
The application reaches the correct Supabase project.
Runtime diagnostics show successful communication with Supabase.
Auth Logs
Supabase Auth Logs show requests such as:
POST /signup → 200
User confirmation requested: request completed
and
POST /token
Email not confirmed
This proves that requests are reaching Supabase.
SQL Verification
Querying auth.users confirmed that users are actually being created.
Querying auth.identities confirmed matching identity records.
Therefore, the backend itself is functioning.
Why I Am Contacting Support
Despite all of the above, AI Studio repeatedly reported successful authentication while the application still failed from a user’s perspective.
Examples include:
- Authentication reported as PASS although Sign In did not function.
- Runtime reports describing behaviour that did not match the actual preview.
- Multiple reports claiming implementation success before manual testing proved otherwise.
- Significant time spent repeatedly debugging issues that later turned out to be unresolved.
The generated implementation reports appeared to rely on expected code paths instead of proving actual runtime behaviour.
What I Need Help With
I would appreciate assistance understanding why AI Studio repeatedly reports successful implementation even though the application behaviour remains incorrect.
Specifically I would like guidance on:
- Whether AI Studio runtime diagnostics are based on actual execution or inferred from source code.
- Whether the Expo preview environment has any known limitations affecting Supabase Authentication.
- Whether AI Studio can incorrectly report PASS without completing a genuine end-to-end authentication test.
- Whether there are known issues with Expo React Native, Supabase Auth, or client-side environment variable injection in AI Studio.
- How to reliably verify that authentication has genuinely been executed instead of inferred.
Impact
This issue has consumed a significant amount of development time because every implementation report suggested that authentication was fully working, while manual testing consistently showed otherwise.
As a result, I have repeatedly debugged the same authentication flow despite receiving multiple PASS reports from AI Studio.
I would appreciate any assistance in identifying the root cause and understanding whether this behaviour is a limitation or bug within AI Studio.
Thank you for your time and support.