Confusing RU limits for /applications write operations — clarification needed

I am implementing automation that creates Azure AD applications and service principals using Microsoft Graph. I am running into frequent 429 Too Many Requests responses, and the documentation around Resource Units (RUs) vs “requests” is unclear.

The documentation states:

“The cost of POST, PATCH, and DELETE operations on the /applications request path depends on the signInAudience type. For apps where the signInAudience is AzureADMyOrg or AzureADMultipleOrgs, the cost is 70,000 requests per 5 minutes.”

Separately, the throttling documentation lists:

Questions

1. Does “70,000 requests per 5 minutes” actually mean 70,000 Resource Units, not HTTP requests?
   The wording suggests HTTP requests, but the throttling behavior suggests RU consumption.

2. Which RU scope applies to the following workflow?

   • GET /applications?$filter=displayName eq 'X'
   • POST /applications (create app registration)
   • POST /servicePrincipals (create Enterprise App)
   • POST /applications/{id}/addPassword (create client secret)

3. What is the RU cost of each of these operations?

4. Is addPassword intended to consume the full 70,000 RU tenant-wide budget?

5. What is the recommended design pattern for automation that needs to create multiple applications, service principals, and secrets without hitting throttling?

Thanks in advance for clarification.

Start by comparing the /applications write-operation wording with the throttling documentation cited in the issue. Document whether the limit is requests or Resource Units, identify the scopes and costs for each listed operation, and describe the recommended automation pattern for avoiding throttling.