A document-processing pipeline used OCR and a language model to pull key fields out of scanned PDFs. Some extracted deadlines were months off. The output always passed validation, so nothing flagged it; a person noticed when a date in the summary did not match the original page.
A document-processing pipeline used OCR and a language model to pull key fields out of scanned PDFs. Some extracted deadlines were months off. The output always passed validation, so nothing flagged it; a person noticed when a date in the summary did not match the original page.
In a scheduling feature, users set a weekly reminder for "Monday at 9:00". Twice a year, right after the clocks changed for daylight saving time, every reminder arrived an hour early or an hour late, then stayed wrong until users edited it.
In a billing-style module, an invoice total sometimes differed by one cent from the sum of its line items. It happened rarely, only for certain quantities and tax rates, and finance users spotted it before any test did.
A third-party CRM-style API had a short outage. It came back within minutes, but our sync service kept failing with 429 Too Many Requests for a long time afterwards, and records stayed out of date.