New submission page is broken

[SkidanovAlex]

When submitting a solution via

https://lean-eval-submission-server.lean-eval.workers.dev/

Pressing on [Submit exact commit] results in:

{
  "error": "provider_unavailable"
}
Document.querySelector: '#' is not a valid selector

The same workflow, with exactly the same fields (except for problem name) worked yesterday.

The exact query to https://lean-eval-submission-server.lean-eval.workers.dev/api/v1/browser/submissions that is being sent from the website (from network page) that results in the error:

{
	"submission": {
		"declared_model": "NEAR AI w/ DeepSeek V4",
		"problem_id": "annals_linear_subspaces",
		"production_metadata": {},
		"publication_choice": "scheduled",
		"schema_version": 2,
		"source_commit": "03e1aecee05cd9a14e7b97140dd96e3f3a826466",
		"source_repository": "SkidanovAlex/leaneval-deepseek",
		"terms_accepted": true,
		"terms_version": "lean-eval-intake-terms-v1"
	}
}

And the response is:

{"error": "provider_unavailable"}

[Vilin97]

Same issue

[kim-em]

The browser's Document.querySelector: '#' is not a valid selector error is fixed and deployed in leanprover/lean-eval-submissions#1925. The form now shows the server's HTTP rejection and response details without throwing that extra exception.

The older provider_unavailable response is still unexplained: it predates the diagnostic fields needed to identify which GitHub operation failed. If this happens again, could you comment here with the UTC time, HTTP status, and the complete response JSON shown by the form (especially request_id, stage, provider_status, provider_operation, and any provider_rate_limit_* or provider_request_id fields)? That will let us trace the failed request. Please share the response, not an OAuth grant or token.

I'm closing this report for the fixed browser error. If a fresh provider failure occurs, please comment here and we'll reopen it to investigate with those diagnostics.