readme: manuals + workflow

README.md

@@ -3,7 +3,10 @@
 10 offline web annotation packs (`annotation_bundles/bundle_01.tar` .. `bundle_10.tar`), 1526 questions in total
 (~150 per pack, stratified by source: fullvideo / excerpt / p01ma_gdrive / p01ma_hf / rbma273).
 
-Each tar contains `review.html`, `questions.json`, `README.md` and `media/` (480p evidence clips).
-Untar, double-click `review.html`, fill in your name, and for every question: watch all clips -> mark every option
-supported / ruled out / undecidable -> tick the final answer -> verdict (clear / ambiguous / cant_tell) -> notes.
-Export the JSON when done and send it back. The packs contain no gold answers.
+Each tar contains `review.html`, `questions.json`, `media/` (480p evidence clips), a `README.md` with the
+step-by-step procedure, and the full judging manual in Chinese (`校验操作手册.md`) and English (`JUDGING_MANUAL.md`).
+The two manuals are also here at the repo root. Read the manual before starting.
+
+Workflow (no blind-guess step): untar -> open `review.html` -> fill in your name -> per question: watch all clips,
+mark every option supported / ruled out / undecidable, tick the final answer, choose a verdict
+(clear / ambiguous / cant_tell), write notes -> Export JSON and send it back. The packs contain no gold answers.