Answered by Falk Gottlob · Reviewed · 1 min read
With a manifest, one row per file, and three counts: files, documents, and the customer's questions that can no longer be answered. Lesson 11 of Becoming Forward Deployed Engineer (November 8, 2026, public section) triages every file with a verdict and a reason, finds 26 unreadable, reports 83 lost documents instead of 26 files, and fixes the denominator at 42 questions after its first scorer let fixed-size chunking score 100% on seven of them. I would change two things. The lesson subtracts four near-duplicate copies from the loss, which it can do because it generated them; in a customer's folder an unread copy stays in the loss, and my own docs agent failed on a current page and its 2021 twin with both readable. And every row gets an owner on the customer's side and the date they were asked. My creator scan reported 0 broken on a run that opened 75 of 149 sources, which is the same silence in a different pipeline.
One row per unreadable file, three counts, and a name on every row that needs the customer. The loss is stated in the questions the customer asked, not in files.
The lesson's triage and its fixed denominator, the twin problem, and the check on my own scan are in Lesson 11 Lost 83 Documents in 26 Files. Count the Questions Too, and Keep the Copies.. The docs agent and its 2021 twin are from Lesson 10 Tripled the Window and Accuracy Held at 60%. Add the Cited Doc and the Twin Column..