OCR with triage, not blind trust
Scanned documents go through OCR and the result is triaged by confidence. Anything above the threshold is accepted automatically; the rest enters a human review queue.
That is the difference between useful OCR and dangerous OCR: a system that accepts everything fills your ledger with misread amounts, and one that accepts nothing saves nobody any time.
Retention, tracked as a date
Every document carries a statutory retention deadline. As it approaches or passes, the document enters a review queue where an administrator decides: extend, place under legal hold, or delete.
There is deliberately no automatic deletion. A document can be past its fiscal retention and still be needed in live litigation — a judgement a cron cannot make.
Attached to what explains it
A document does not sit alone: it is attached to the invoice, contract or partner it belongs to. Search finds it by content, not only by filename.
Worth knowing
An archive can fail in two directions: deleting what should have been kept, and keeping forever what should have been deleted. The second looks harmless but is equally a breach under data-protection law — which is why retention is a tracked date rather than an intention.