Published Updated

Confirm the accepted content before calling a large upload complete. A progress bar describes the client's transfer; the application also needs evidence that the intended bytes reached storage and are associated with the correct document. Required inspection and processing have their own outcomes.
Consider an inspection report whose transfer reaches S3 just before the connection fails. The client never receives the success response. Immediately creating another upload can duplicate work or create competing writes. Keep the outcome pending while the integration checks what arrived, then resume the unfinished work.
Use the supported upload path and retain the expected content identity with the request.
Reconcile acceptance, integrity, and the document reference before deciding whether another transfer is needed.
Track inspection, extraction, indexing, and business release independently of upload progress.
A useful interface can say “confirming upload,” “processing,” “ready,” or “action required.” These are example application labels. Define the evidence behind each one so a timeout or unavailable scanner leaves a visible pending or exception state.
Retain an integration-owned operation reference, site, business case, intended document, and whether the request adds content or replaces it.
A deliberate corrected submission is a new business decision, even if its filename is unchanged.
Track transfer attempts, expected length and checksum, accepted content references, observed results, and processing attempts. Retain version identifiers where the selected deployment supports document versioning.
Keep presigned URLs and credentials out of routine logs; use opaque correlation references.
The operation reference belongs to the integration's recovery record. It is not an assumed Idempotency-Key header on the FormKiQ API. Use documented request fields and retrieval operations, and retain any additional correlation state in the application or integration.
FormKiQ's checksum guidance shows how to supply a checksum and algorithm when requesting the upload URL. Send the returned checksum headers exactly as provided. The request representation and the S3 header representation can differ; do not substitute a displayed hexadecimal digest for a returned header value.
Compare the expected and observed length, algorithm, digest, and content reference. A matching digest helps establish that the accepted bytes match the declared file. It does not decide who may submit the file, whether it contains malware, or whether it satisfies the business requirement.
For a custom multipart integration, record the checksum type as well as the algorithm. A composite checksum is calculated differently from a full-file checksum. An S3 ETag is also not a universal full-file digest. Use the applicable S3 integrity rules when comparing results.
| Milestone | Evidence to check |
|---|---|
| Upload requested | The intended site and document are identified and temporary upload authority has been issued. |
| Transfer confirmed | The storage operation succeeded, or its uncertain outcome has been reconciled with the accepted content. |
| Record reconciled | The intended length and checksum match the content associated with the document; retain its version reference where available. |
| Processing assessed | Required actions have recorded outcomes for that content. Pending, failed, unsupported, and successful results remain distinguishable. |
| Ready for the next task | The configured inspection and business conditions permit release to the intended audience or workflow. |
Upload capacity and processing capacity are different limits. Check the supported format, file size, execution limits, and configured rules for each selected action or module. A successfully stored file may need an exception route if it cannot pass the required processing. For the queueing side of this design, see document ingestion and backpressure.
After a lost response, check the original document and available acceptance evidence before requesting another upload. If the intended content is already present, continue reconciliation and processing. If acceptance remains uncertain, keep the operation pending for recovery. Retry transfer only under a defined rule that accounts for expiry and competing writes.
AWS documents that a presigned URL can be reused until it expires, and another write to the same key can replace current content. A correlation ID in a database cannot by itself stop an older valid URL from writing. Coordinate replacements through the supported upload flow; where concurrent attempts require isolation, use a deliberately designed staging and handoff integration. Verify how the chosen design prevents a late attempt from changing content already accepted for review.
Bind processing to the accepted content reference and use duplicate-safe workers and downstream operations. S3 notifications use at-least-once delivery, so one notification is not a guarantee of one execution. Keep the business outcome and individual attempts linked, as in the document-action retry guide.
This fictional integration test uses a 250 MB inspection report. The file is within the limits chosen for the test deployment's processing path; it is not a claim that every module accepts that size. Operation UPL-00041 concerns one document in the engineering site.
The application records the source reference, expected length and SHA-256 digest, intended site, and returned document identifier. The user sees an upload in progress. Issuing the URL does not mark the inspection report ready.
The transfer reaches S3, but the client connection fails before confirmation arrives. The application changes the display to “confirming upload” and retains the original operation reference. It does not immediately create a second document.
The recovery process checks the document and the available storage or ingestion evidence under its authorised identity. The expected content reference, length, and checksum match. It records the confirmed result and the relevant version reference where versioning is enabled. A mismatch remains an exception.
Two notifications reach the processing path. Its deduplication and recovery rules associate both with the same accepted content. Each required action records its own result; retries resume the unfinished work without creating a second business submission.
When the required checks succeed, the application allows the next review task. A failed inspection, unsupported file, or processing timeout stays visible with an owner and recovery action. The transfer outcome remains confirmed while that separate issue is resolved.
The documented FormKiQ presigned PUT flow is separate from a custom S3 multipart integration. For multipart delivery, confirm successful assembly of the parts before treating the object as complete. AWS notes that CompleteMultipartUpload can return an embedded error after an initial HTTP 200 response; use the SDK's error handling or inspect the complete response.
Configure AbortIncompleteMultipartUpload lifecycle rules for abandoned multipart parts. This action does not delete completed objects. Separately expire unused application sessions and resolve their reservations, while preserving the evidence needed to explain the outcome. A completed document follows its own retention and hold rules.
A reliable upload path confirms the intended bytes, connects them to the right document, and reports the required processing separately. When a response is lost, reconciliation gives the application a basis for continuing safely. That evidence makes the difference between a completed transfer and a document ready for business use.
The open-source foundation — API-first, deployable into your own AWS account, and free to use. Right for architecture validation and early implementation.
Production-ready editions for departments and complex workflows. Start with a Proof-of-Value deployment or go straight to production.
For governance-heavy environments with residency, sovereignty, assurance, and multi-jurisdiction requirements. Talk to us about the right deployment model.