Troubleshooting

Start by identifying where the problem occurred: signing in, preparing input, submitting the request, waiting for a batch, or interpreting a completed result. The same visible symptom can have different causes, and repeated submissions can consume quota without resolving them.

I cannot sign in

What you see	What to do
The code email has not arrived	Check the entered address and spam or junk folder. Wait for the displayed resend interval before requesting another code.
The code is wrong or expired	Enter the six digits in the browser that requested them. Codes expire after ten minutes and have a limited number of attempts; request a fresh code if needed.
No code was requested	The matching browser challenge is missing. Request a code in the browser or private session where you intend to sign in.
Too many sign-in attempts	Honor the retry interval. Multiple tabs or repeated sends do not bypass the account, IP or service-wide limits.
The console says the session has ended	Sign in again with the same email account. Preserve an existing audit id before navigating away.

The account guide explains how sign-in, sessions and API keys fit together.

The console or API rejects my input

Use the field named in the error to find the smallest correction. A JSON request can be syntactically valid and still fail the selected use case's schema.

Amount rejected: use a decimal string in JSON, such as "212.40". Do not add a currency symbol or a thousands separator to the value.
Date rejected: use the field's published format. A date is YYYY-MM-DD; fields requiring a date-time also need the expected time and offset.
Unknown field or item type: confirm the selected use case and compare the names with its published schema.
Missing required field: pasted JSON needs every required field. For an imported file, map each required field to a column or a constant; a row whose required cell is blank is withheld, and the preview lists it with its reason.
Repeated item id: each type-and-id pair in a batch must be unique. This request validation is separate from a duplicate-payment finding.
Request too large: count the encoded items and context together. Reducing the items alone may not reduce a large shared context enough. An imported file is at most 32 MiB.

For an imported file, select Preview the mapping and read each withheld row's reason: a ragged row, a blank required cell or a value that does not fit its field. Remove repeated header names, export dates as text, and if no row is admitted, the run is refused until the preview admits one. A busy file reader or full storage asks you to retry after the stated seconds. Read the console guide and schema guide for the expected input.

The request succeeded, but the verdict is uncertain

uncertain is a completed result. Read undecided and the associated assertions to decide what should happen next.

Reason	Useful next step
`insufficient_evidence`	Inspect the named check and provide the missing record, reference or source passage. Do not replace an unknown comparison population with an empty list just to obtain a result.
`unqualified`	The model question cannot decide under the current qualification. Route it to review; a high or low recorded probability does not change that status.
`abstained`	The model answer falls outside the accepted decision regions. Review the evidence and resolve the underlying business question.
`guard`	Review indicators reached the configured guard. Examine their reasons alongside the assertions.
`combined_bound`	Individual model answers hold, but their combined error bounds exceed the gate budget. Retain the uncertain result for review; lowering the target changes the decision policy, not the evidence.
`nothing_applicable`	No assertion remains applicable after scope exclusions. Review the selected item type, applicability and any explicit skips; nothing tested is not a pass.
`not_asked` or unavailable model evidence	Read the assertion's message and the request options before rerunning. Determine whether the check was deliberately omitted or could not execute.

Use Reading results to distinguish the verdict, probability, certainty and risk. The evaluations explain what measurements are available.

The result is false, but the request returned HTTP 200

That is expected. HTTP 200 means the verification completed; false means a failure satisfies the gate's exception rule. In the supplier-invoice example, po_present fails because the $24,800.00 invoice has no PO reference above the $1,000.00 threshold. Obtain supporting evidence or document the control exception with its owner. Resubmitting unchanged evidence is not a way to resolve the finding.

A false verdict does not mean every assertion was decided. Its undecided list can be empty while individual assertions remain unresolved or unasked, so inspect assertions before deciding which follow-up work is complete.

If the input was wrong, correct it in the source workflow and rerun the relevant check. If you disagree with a completed audit's finding, preserve the evidence and use the review-marks API to record the review. Marks do not automatically change model thresholds or retrain the service.

The certainty is below my target, or is unmeasured

The requested target controls the gate's error budget. Returned certainty is a separate lookup of published evaluation evidence for the verdict group; it does not veto a verdict when its value is below the target. An uncertain verdict has null certainty. A decided result can instead report basis: "unmeasured" when no applicable table entry exists. Read the basis, bound and counts rather than treating the target as an accuracy guarantee. The certainty investigation shows recorded examples of these distinctions.

Does materiality change a failed result?

The current API does not accept a materiality setting or waive assertions based on one. A reviewer assesses materiality, the nature of the exception and any further testing within the engagement. Keep that judgment and the resolution alongside the returned finding in the audit trail. Read Audit concepts for the distinction between a control exception, substantive testing and the engine's verdict.

A batch appears stuck or its results are missing

A submitted audit moves through queued and running states before completing or failing. The console displays its id and progress. For a known audit id, the batch API lets you check its current status and retrieve its results when complete.

If the state is queued or running, inspect progress and allow time for processing.
If it is failed, read the audit error and correct the cause before a deliberate resubmission.
If reading items returns not_completed, wait for a completed state rather than repeatedly requesting items.
If an audit returns not_found, check the id, owning account, deletion and retention window.
If a submission timed out before you received an id, do not assume it was rejected. Repeat it with the same Idempotency-Key and body to get the audit it created; without a key, repeating the POST may create another audit.

Reloading or closing the console does not preserve its on-screen state. Download results and keep the audit id during the review. Completed audit results expire after seven days unless deleted sooner.

I reached a limit or my API key stopped working

For rate_limited, wait for Retry-After and reduce concurrent requests from the same account. Daily item or question quotas reset at 00:00 UTC; retrieving existing results remains possible within the rolling request limit. Check the account page and limits guide.

For invalid_api_key, confirm that the integration is sending the complete current key. It may have expired or been revoked. Replace a lost key from the account page rather than trying to retrieve its secret value. The authentication guide explains the supported credentials.

The error reference covers HTTP codes, service availability and timeout handling in more detail.

Prepare a useful problem report

Use the ERP.AI contact page for an unresolved issue and identify OpenAuditor as the product. Include:

The page or API endpoint and what you were trying to do.
The approximate time and time zone, plus the browser or client used.
The exact visible error, HTTP status and error code when available.
The Request-Id from the API response header, or the audit id shown by the console.
A short sequence of steps and the smallest sanitized example that reproduces the problem.

Leave out API keys, sign-in codes, session cookies and confidential customer records. Check screenshots for account details and source data before sharing them. For a security concern, use the disclosure channel listed on the same contact page and consult OpenAuditor's security information.