What evidence would make you trust an AI gateway's fallback charge?

I work on TackleKey, an OpenAI-compatible AI API gateway.

The product question we are testing is narrower than “which gateway has the lowest model price?”

Suppose you request model A through provider 1. Provider 1 times out, the gateway falls back to provider 2, and the response succeeds. What evidence would make you trust the final charge?

Our current answer is a request receipt that separates:

• requested model;

• preferred route;

• final serving source;

• retry/fallback path;

• billable usage;

• final user charge.

We also test that one logical request does not create two user-visible charges, and that the price shown in logs comes from the source that actually completed the request.

The smallest useful validation is:

1. one successful request;

2. one controlled failure/fallback;

3. one balance reconciliation;

4. one small quote-to-checkout check.

Here is the current team-oriented checklist:

https://tacklekey.com/solutions/team-ai-api-cost-governance?utm_source=producthunt&utm_medium=discussion_reply&utm_campaign=company_signal_proof&utm_content=fallback-charge-trust-question-20260729

The feedback question: if you manage AI spend for a product or team, which part of that receipt is still missing before you would put real traffic or money through the gateway?