第四課：選對分析、選對 PersonaChoosing the Right Analysis and Persona
為什麼這件事會出錯Why this goes wrong
有位同學在 jamovi Module Guider（askllm）裡問：「幫我寫一段跑獨立樣本 t 檢定的 R 程式碼」，結果沒拿到程式碼，只看到回覆說要改用另一個分析。他以為 工具壞了，其實是踩到了 askLLM 設計裡的一條硬規則：Module Guider 的定位是 推薦 jamovi GUI 分析、逐字引用選單路徑，不是 R 產碼助理；只要問題涉及 R 程式碼，它一定會把你導去 Analyses ▸ askLLM ▸ R code tutor，不論你選的是 哪個 persona、哪個語言。反過來，如果在 R code tutor（askllmr）裡問 「這筆資料該用哪個 jamovi 選單」，它也一定會請你改用「jamovi Module Guider」，而不是自己亂猜選單路徑。這是雙向、寫死、不能關閉的邊界句。
A student asked the jamovi Module Guider (askllm), “write me R code for an independent-samples t-test.” No code came back. The reply simply told her to use a different analysis. She assumed the tool was broken. In fact she had hit a hard rule built into askLLM: the Module Guider’s job is to recommend a jamovi GUI analysis and quote menu paths verbatim, not to write R code. Whenever a question asks for R code, it always redirects to Analyses ▸ askLLM ▸ R code tutor, no matter which persona or language is selected. The reverse holds too. Ask the R code tutor (askllmr) which jamovi menu to use, and it will tell you to switch to the “jamovi Module Guider” instead of guessing a menu path itself. This boundary runs both ways, is hard-coded, and cannot be turned off.
兩種分析怎麼分工How the two analyses divide the work
一個模組、兩個分析：askllm（jamovi Module Guider，選單標題就是「jamovi Module Guider」）負責「該跑哪個分析、選單路徑是什麼」；askllmr（R code tutor）負責「這段邏輯的 R 程式碼怎麼寫，貼進 Rj Editor 能跑」。兩者共用同一 套摘要統計、同一組 persona，但角色互斥且互相導流：問錯棚會被請去對的棚，不 會得到將就的答案。開啟 Rj 的路徑是 Analyses ▸ R ▸ Rj ▸ Rj Editor；若本機 沒裝 Rj，R code tutor 的結果面板會提示 Modules ▸ jamovi library 去安裝。
One module, two analyses. askllm (menu title “jamovi Module Guider”) answers “which analysis, which menu path.” askllmr (R code tutor) answers “what R code implements this, ready to paste into the Rj Editor.” Both share the same summary statistics and the same personas, but their roles are mutually exclusive. Ask the wrong one and you get redirected to the right one, not a half-answer. Rj Editor opens at Analyses ▸ R ▸ Rj ▸ Rj Editor. Without Rj installed, the R code tutor’s result panel points you to Modules ▸ jamovi library to install it.
三種 Persona 怎麼選Choosing among the three personas
Persona 選項（role）有三個值：consultant（🧭 顧問）、tutor（🎓 導師）、explainer（💡 解說員），每個都有獨立的 en/zh system prompt 模 板。consultant 的設定是「精簡、準確、務實」，直接給結論；tutor 採「蘇格拉 底式教學」，不直接給答案，而是用提問引導你自己推理；explainer 假設你完全 沒有統計背景，逐步解釋並定義每個統計名詞。這三者不分優劣，只看你要的是效 率還是理解。下表是情境對照：
The persona option (role) has three values: consultant (🧭), tutor (🎓), and explainer (💡), each with its own English/Chinese system-prompt template. Consultant is built to be “concise, accurate, practical” and gives you a conclusion directly. Tutor uses “Socratic dialogue.” It will not give the final answer, but asks guiding questions so you reason it out yourself. Explainer assumes no statistics background at all and defines every term it uses. None is objectively better. The choice depends on whether you want speed or understanding. The table below maps scenarios to choices:
| 情境 | 選哪個分析 | 選哪個 persona |
|---|---|---|
| 不知道該跑哪個分析，想要選單路徑 | Module Guider | consultant（快速拿結論） |
| 已經知道分析，想理解「為什麼是這個」 | Module Guider | tutor |
| 完全零基礎，需要每個統計名詞被解釋 | Module Guider 或 R code tutor | explainer |
| 需要貼進 Rj Editor 就能跑的 R 程式碼 | R code tutor | consultant 或 explainer（視背景） |
| 課堂上教零基礎學生寫 R | R code tutor | explainer |
| 想要蘇格拉底式自學，不想被直接餵答案 | Module Guider 或 R code tutor | tutor |
| Scenario | Which analysis | Which persona |
|---|---|---|
| Don’t know which analysis to run, want a menu path | Module Guider | consultant (fast answer) |
| Know the analysis, want to understand “why this one” | Module Guider | tutor |
| Complete beginner, needs every term defined | Module Guider or R code tutor | explainer |
| Need R code ready to paste into Rj Editor | R code tutor | consultant or explainer (depending on background) |
| Teaching R to beginners in class | R code tutor | explainer |
| Want Socratic self-study, not a direct answer | Module Guider or R code tutor | tutor |
進階：用變數 Description 自訂 system promptAdvanced: a variable’s Description as a custom system prompt
兩個分析都有一個選項 systemPromptVar，標題是「Use a variable’s Description as the system prompt」，型別是 Variable，預設 NULL。用法：在資料裡選一個 變數，到它的 Setup 面板填入 Description 文字，再到 systemPromptVar 選項 選那個變數——askLLM 會把這段 Description 文字當成自訂 system prompt，取代 原本 consultant/tutor/explainer 的模板（優先序是「自訂內容非空 > persona 模板」）。這讓你（或整班）可以固定一套教學語氣或角色設定，不必每次都重寫 提問。要提醒的是，這是 askLLM 讀取變數 metadata 的一種設計，並非 jamovi 官 方文件逐條記載的通用機制，若行為與這裡描述不同，請以你實際測試到的為準。
Both analyses share an option called systemPromptVar, titled “Use a variable’s Description as the system prompt,” of type Variable, defaulting to NULL. To use it: pick a variable in your data, open its Setup panel, type your text into the Description field, then select that variable in the systemPromptVar option. askLLM will use that Description text as a custom system prompt, replacing the built-in consultant/tutor/explainer template. Non-empty custom text always takes priority over the persona template. This lets you, or a whole class, fix one teaching voice or role without retyping it into every question. One caveat: this is a design choice in how askLLM reads variable metadata, not a generic mechanism documented in jamovi’s own manuals. Trust what you actually observe over this description when your installation behaves differently.
動手做（5 分鐘）Try it (5 minutes)
- 在 Module Guider 裡，用預設的 consultant persona 問一個統計問題（例如 「這兩組的平均數有差異嗎？」），記下回覆的語氣與長度。
- 把 persona 改成 tutor，問一模一樣的問題（記得先取消再重新勾選 Submit），比較這次回覆是不是變成提問而不是直接給答案。
- 在 Module Guider 裡問一句要求 R 程式碼的問題，確認回覆是不是把你導向
Analyses ▸ askLLM ▸ R code tutor，而不是直接生成程式碼。 - 找一個變數，在 Setup 面板的 Description 欄填一句系統提示（例如「請永遠 用一句話總結」），到
systemPromptVar選它，重問一次同一個問題，觀察 語氣是否照你寫的規則改變。
- In the Module Guider, ask a statistics question (e.g. “is there a difference between the two group means?”) with the default consultant persona, and note the tone and length of the reply.
- Switch the persona to tutor and ask the exact same question (untick and re-tick Submit first), and check whether the reply now asks you questions instead of giving a direct answer.
- In the Module Guider, ask for R code and confirm the reply redirects you to
Analyses ▸ askLLM ▸ R code tutorinstead of generating code directly. - Pick a variable, type a one-line system prompt into its Description field (e.g. “always summarize in one sentence”), select it in
systemPromptVar, ask the same question again, and see whether the tone follows your instruction.
帶走這句話One thing to remember
分析選錯棚、persona 選錯調性，askLLM 會把你導回正確的分析，但語氣與深度永 遠要你自己配對到情境。
Pick the wrong analysis or the wrong persona tone, and askLLM will redirect you to the right analysis. But matching tone and depth to your situation is always on you.