ChinaAI: Kimi consumer PDF reading test records

Observed 23 September 2026 UTC in the signed-in official kimi.ai web chat.
One first answer per PDF, each in a fresh chat, using the visible Instant High
selector. The serving checkpoint was not exposed. No regeneration or corrective
follow-up. The supply-order response used Kimi's built-in Python tool within
its first answer. No ChinaAI visitor chat or developer API call was made.

These source files are fictional. They were reused unchanged from earlier
ChinaAI PDF guides, but the Kimi answers here were newly collected. Different
services, dates and settings do not make this a matched comparison.

Read protocol.md for the pre-request test conditions and success checks;
test-results.md for observed outcomes; evidence/*-first-answer.md for the
complete final replies. The screenshots show actual UI states. Do not infer
an accuracy percentage or general performance from one attempt per file.

No private chat URL, account identifier, key or unrelated draft is included.
