技術細節版 · 系統與資料流 — 從錄音到出版，平台校對為主線Technical version · systems & data flow — from recording to publication, platform review as the main line
階段一Stage 1✓ 已自動化✓ Automated
錄音 → 逐字稿Recording → Transcript
已有自動化腳本完成（gcdr-community/oseling-auto-transcript），技術端不需人工。Handled by an automated script (gcdr-community/oseling-auto-transcript); no manual work.
STEP 1
🎧 現場錄音🎧 On-site recording
開示現場錄音檔（英文開示＋中文口譯）Live recording of the teaching (English + Chinese interpretation)
STEP 2
🤖 AI 逐字稿🤖 AI transcript
腳本用 AI（Gemini）自動轉錄成逐字稿，附時間戳A script auto-transcribes with AI (Gemini), with timestamps
STEP 3
✍ 初步編輯對稿✍ First-pass editing
自動潤飾英文、整理中文，並三段對齊Auto-polishes English, organizes Chinese, aligns the three parts
產出三段：交錯時間戳稿 · 英文潤飾稿 · 中文整理稿 → 交給下一階段合成三欄校對稿Outputs three parts: timestamped transcript · polished English · organized Chinese → passed to the next stage to build the three-column review doc
▼
階段二Stage 2
產製 · 做出「可校對的文稿」Production · make the reviewable document
承接上面三段稿，技術端一次性處理，之後不再重建（重建會蓋掉志工校對）。Takes the three parts above, processed once; not rebuilt afterwards (rebuilding would overwrite volunteer edits).
STEP 1
合成 masterBuild master
build_master.py：三欄對照 🎙原音 / 🇬🇧英校 / 🇹🇼中校，英文骨架對齊、整句分配build_master.py: three columns 🎙audio / 🇬🇧English / 🇹🇼Chinese, aligned on the English skeleton, whole-sentence distribution
STEP 2
音檔 + 基準Audio + baseline
音檔上 R2（faststart 可 seek）；AI 原始稿存 .baseline（供逐字標色比對）Audio to R2 (faststart, seekable); the AI draft saved as .baseline (for word-level change highlighting)
志工用 Email 登入（Cloudflare Access 白名單），無需主機、無需 GitHub 帳號。存檔即 git commit。Volunteers log in by email (Cloudflare Access whitelist); no server, no GitHub account. Saving = a git commit.
① 登入 · 看指派① Log in · assignments
首頁只顯示指派給自己的文稿，附校對／審定進度條Home shows only your assigned docs, with proofread/verify progress bars
→
② 逐段校對② Review by paragraph
聽英文原音 → 核對文意與詞彙 → 只勾自己有權限的中/英校對·審定Listen to the English → check meaning & terms → tick only the ZH/EN proofread·verify you’re allowed
→
③ 存檔③ Save
逐字上色顯示改動 → 按存檔（= 一次 commit，可回溯）Word-level highlighting of changes → Save (= one commit, reversible)
📝 校對📝 Proofread
確認字義、句意正確：意思忠於英文、名相與詞彙對、無誤譯漏譯Confirm word & sentence meaning: faithful to the English, correct terms, no mistranslation/omission
✅ 審定✅ Verify
在校對之上，確認文句流暢、文字品質，達可閱讀與出版水準Beyond proofreading: confirm fluency and quality, ready to read and publish
逐字標色：琥珀＝相對 AI 原稿已修改綠＝你剛改 · 每段可「↺ 恢復」回打開時的樣子 · 🕘 編輯歷史看誰何時改了什麼 · 段落鎖一次一人，可強制接手Word highlighting：amber = changed vs the AI draftgreen = your edits · “↺ Revert” each paragraph to its reopened state · 🕘 Edit history shows who changed what · Paragraph lock, one at a time, force-takeover available
▼
階段四Stage 4
完稿與衍生Final text & outputs
校對＋審定完成 → 定稿 master（單一真相），再依用途衍生產出。Proofread + verified → final master (single source of truth), then derive outputs by use.
🌐 官網法教資料庫🌐 Website archive
放完整版一篇（單一真相）One full version (single source of truth)
📖 出版📖 Publishing
審定稿供編輯、排版、成書Verified text for editing, layout, book
🚧 下一步（規劃中）· 社群自動發佈平台：AI 摘錄（從定稿自動生成貼文草稿）→ 人工審稿 → 腳本自動發佈到 Facebook · IG · LINE🚧 Next (planned) · social auto-publish: AI excerpting (auto-draft posts from the final text) → human review → script auto-publishes to Facebook · IG · LINE
🏢 全部建置在組織帳號下 · gcdr-community（社群自持，不依賴個人帳號）🏢 All built under the org’s own accounts · gcdr-community (community-owned, not personal)
🐙 GitHub 組織 gcdr-community🐙 GitHub org gcdr-community所有程式碼與內容都放這（皆 private）：逐字稿腳本、build_master、校對平台 App、教導 vault（存 master 與權限）。存檔即 git commit、可回溯。All code & content here (all private): transcript scripts, build_master, the review app, the teachings vault (master + permissions). Saving = a git commit, reversible.
📁 Google Drive（組織）📁 Google Drive (org)開示錄音檔的來源；逐字稿管線從這裡取檔。Source of the recordings; the transcription pipeline reads from here.