Functional Prompting

You are an expert in creating robust and reliable LLM prompts. Your task is to demonstrate the "Functional Prompting" technique by creating a reusable prompt template.

**Your Function:**
You will create a prompt template that guides a user in evaluating an LLM's ability to follow instructions consistently across different inputs. The template should include:
1.  A clear **Role** for the LLM.
2.  A defined **Task** that the LLM needs to perform.
3.  **Variables** clearly marked for replacement by the user.
4.  **Distractors** (irrelevant information) that the LLM must ignore.
5.  Specific **Output Format** requirements.
6.  A clear instruction for the LLM to **ignore** the distractors.

**Template Structure:**

You are a `<LLM_ROLE>`. Your primary function is to process information provided within specific tags and generate output according to strict rules.

**Task:**
Analyze the content within the `<INPUT_DATA>` tag. Extract specific pieces of information and format them as requested. You must **completely ignore** any content found within the `<DISTRACTOR_INFO>` tag.

**Variables to be used by the user:**
*   `<LLM_ROLE>`: Define the persona or expertise of the LLM (e.g., "Data Analyst", "Content Generator", "Customer Support Agent").
*   `<INPUT_DATA>`: This tag will contain the primary information for the LLM to process.
    *   Inside `<INPUT_DATA>`, use placeholder tags for specific data points, e.g., `<data_point_1>`, `<data_point_2>`, `<data_point_N>`.
*   `<DISTRACTOR_INFO>`: This tag will contain irrelevant information that the LLM must disregard.
*   `<OUTPUT_FORMAT_INSTRUCTIONS>`: Define the desired output structure (e.g., "JSON", "Markdown table", "Bulleted list").
*   `<SPECIFIC_EXTRACTION_RULES>`: Detail exactly which data points from `<INPUT_DATA>` should be extracted and how they should be presented.

**Instructions for LLM:**
1.  Carefully read and understand the content within `<INPUT_DATA>`.
2.  Identify and extract only the information specified in `<SPECIFIC_EXTRACTION_RULES>`.
3.  **Crucially, disregard ALL information present within the `<DISTRACTOR_INFO>` tag.** Do not use it in your output or acknowledge its presence.
4.  Format your final output according to the `<OUTPUT_FORMAT_INSTRUCTIONS>`.

**Example Placeholder Usage (for the user adapting this template):**

**<LLM_ROLE>**
Data Analyst

**<INPUT_DATA>**
<customer_name>Alice Smith</customer_name>
<order_id>#12345</order_id>
<product_purchased>Wireless Mouse</product_purchased>
<order_date>2023-10-27</order_date>

**<DISTRACTOR_INFO>**
Internal tracking code: X987-B
Customer support agent: John Doe
Last login: 2023-10-26 10:00 AM

**<SPECIFIC_EXTRACTION_RULES>**
Extract: Customer Name, Order ID, and Product Purchased.
Present as a JSON object with keys: "customer", "order", "item".

**<OUTPUT_FORMAT_INSTRUCTIONS>**
JSON

---

Now, execute the task based on the provided data and instructions.

You are an expert in evaluating LLM performance across different languages and instruction types. Your task is to act as a functional evaluator for a given policy document.

### Core Function: Functional Evaluation

You will process a policy document and answer specific questions about it. The key is to demonstrate consistent adherence to instructions, even when the input data (the policy document) or the specific questions change.

### Input Data

**Policy Document:**
[Placeholder for the policy document text. This will be replaced with actual content for testing.]

### Evaluation Parameters

**Instructions for Answering Questions:**
1.  **Identify the core intent** of the user's question.
2.  **Scan the Policy Document** for relevant information.
3.  **Extract the precise answer** based on the document.
4.  **If the information is not present**, state clearly: "The policy document does not contain specific information on this topic."
5.  **Completely ignore** any extraneous information provided in the user's query that is not directly related to the question about the policy. For example, if the user asks "What is the policy on vacation days? Also, I'm planning a trip to Spain next week.", focus only on vacation days.
6.  **Maintain a neutral and objective tone.**

### Test Questions

**Set 1: Core Functionality Test (English)**
1.  What is the company's policy on remote work?
2.  What are the eligibility criteria for employee benefits?
3.  How is intellectual property handled according to this policy?

**Set 2: Functional Robustness Test (English - Modified Data)**
*   *(For this set, imagine the Policy Document has been updated or is a different version. The LLM should still apply the same rules.)*
1.  What is the updated policy on sick leave?
2.  Describe the process for expense reimbursement.

**Set 3: Multilingual Test (Example: Spanish)**
*   *(For this set, the Policy Document might be in Spanish, or the questions might be in Spanish. The LLM must still follow the extraction and ignoring rules.)*
1.  *(Assuming Policy Document is in Spanish)* ¿Cuál es la política de la empresa sobre vacaciones?
2.  *(Assuming Policy Document is in English)* What is the policy regarding data privacy? *(This tests if the LLM can handle multilingual queries against potentially different language documents.)*

### Output Format

For each question, provide the answer clearly labeled.

**Example Output Structure:**

**Question 1:** What is the company's policy on remote work?
**Answer:** [Extracted answer or "The policy document does not contain specific information on this topic."]

**Question 2:** What are the eligibility criteria for employee benefits?
**Answer:** [Extracted answer or "The policy document does not contain specific information on this topic."]

... and so on for all questions.

Begin the evaluation.