"Change Applied" or Broken Code? The token loss loop due to syntax errors in the runtime layer

The Problem: The Syntax Error Token Loop

This isn't about the AI misunderstanding a complex instruction (which is a normal part of an iterative process). The critical issue occurs when the platform confirms a change as "Applied" and consumes full credits, yet injects elementary syntax errors (for example, unclosed adjacent JSX elements in any jsx with problem generated by AI).

🫤 This creates an unacceptable financial loop:

• The AI introduces a basic compilation error.

• The system charges tokens for that broken code.

• The user must spend even more of their own tokens trying to get the AI to fix the very error it just introduced.

Execution Layer Failure vs. AI Hallucination

Base44 support argues that "AI errors are an inherent part of how the technology functions." However, delivering a file that breaks the application build while marking it as successful is not a creative hallucination; it is a failure in the platform's validation and execution layer. If the code is not syntactically valid, the system should not consider it "applied" or charge the full code-writing rate.

Lack of Transparency (Log Auditing)

When requesting a technical review of the generation/apply logs to prove that code was not updating correctly despite being charged, support's response was a flat refusal to provide internal logs. For professional or regulated development environments, working blindly without traceability regarding what the platform actually writes versus what it charges represents a massive operational risk.

Call to the Community

• How many of you here have lost dozens of credits just trying to revert your app back to a functional state after an AI commit broke the compiler?

• Shouldn't Base44 implement a pre-apply linter or validator that halts token consumption if the generated code destroys the application build?

• Is it fair for the user to fund the computational cost of syntax errors introduced by the platform itself?