ACTUAL MULTI-DOCUMENT QUESTION/ANSWER EXPORTS - 2026-10-12

cedar-notes-a.pdf and cedar-notes-b.pdf are two self-authored one-page sources with fictional tray records. They are not field measurements or care instructions. Source A deliberately has no temperature measurement; both sources omit growth outcomes and a control group.
question-1.txt and question-2.txt contain the exact submitted questions. One conversation used the same two sources for two completed Qwen 3.8 Flash questions. No answer text was edited and no additional question was submitted for export.

qa-first-default.md: native first-round export with question and answer, without the Source evidence option. Inline citation numbers remain, but the source filenames and appendix do not.
qa-first-evidence.md: the same question and answer with source evidence checked. Thirteen numbered source items are appended. This isolates the evidence option; it is not a comparison of two different AI answers.
qa-second-evidence.md: only the original conversation's second round, with question, answer and evidence. Its exported heading is re-numbered to "第 1 轮". Confirm which question is present, rather than assuming the export heading preserves the conversation index.
qa-both-evidence.docx: native Word export containing both rounds, their questions, answers and source evidence. It has five native tables, no Office Math objects, no Word footnote references or hyperlinks, and no embedded source PDFs. It has not been opened or edited in Word/WPS for this check.
checks.json: hashes, source quantities, checks and limitations.

The workspace shows source filenames, page 1 and source-range highlights. The Markdown and Word exports retain filenames and excerpts but do not include page labels in this example. Keep the separate source PDFs and record page numbers when needed. Inline citation numbers are not a formal bibliography or a complete original-document archive.
The first response's 11 visible evidence groups become 13 separate export items because 6-7 and 8-9 are grouped in the UI. The second round has 8 items. Citation numbers are local to their answer.
The calculation 450/6 = 600/8 = 75 is a per-tray daily average. The follow-up calculation 450/18 = 600/24 = 25 is a per-seedling daily average, not evidence that each individual actually received that quantity. Different observation periods and absent growth data do not establish which set grew better.
All six source/native files are preserved byte-for-byte. This is an inspectable example, not an accuracy benchmark. Keep the original downloads before making further edits.
