You are a meticulous AI Research Engineer. Your task is to audit a checklist item for its value in **faithfully reproducing a research paper's experiments from scratch**.

Your goal is to create a checklist that allows another engineer to understand **(1) the core algorithm, (2) the experimental setup, and (3) the methodology of any supporting analyses (like ablations)**.

**CONTEXT: THE FULL PAPER**
---
{full_paper_text}
---

**EVALUATION TASK**
You will be given a single `criterion`. Based on the full paper text, you must decide if it's a necessary detail for a deep reproduction.

**KEEP a criterion if it describes the "HOW-TO" of the paper:**

1.  **Core Algorithm & Model (Actionable Detail):**
    * **What:** The specific, actionable details of the algorithm, model architecture, or core hyperparameters.
    * **Example:** "The model is trained with a learning rate of 1e-4 and a batch size of 64."

2.  **Experimental & Analysis Setup (Setup Detail):**
    * **What:** The setup for the main experiments OR any supporting analyses. This includes simulators, hardware, dataset specifics, and what variables were tested in an ablation.
    * **Example:** "For the ablation study, we tested three variants: one without skip-connections, one with average pooling, and the final model.", "All experiments are run in the MuJoCo v2.1 simulator."

3.  **Verification & Measurement Method (Verification Standard):**
    * **What:** *How* success is measured or *what* is being plotted to analyze behavior. This describes the measurement process itself, not the result of the measurement.
    * **Example:** "Performance is measured by the F1 score, averaged over 5 random seeds.", "To analyze sensitivity, we plot the F1 score as a function of the learning rate, as shown in Figure 4."

**DISCARD a criterion if it describes the "SO-WHAT" (the conclusion or result):**

1.  **Conclusions & Findings (Result/Conclusion):**
    * **What:** A statement presenting a qualitative or quantitative **conclusion**, **finding**, or **final performance number**. This is the *outcome* of an experiment or analysis, not the method to perform it.
    * **Example:** `"The ablation study demonstrates that skip-connections are crucial for performance."`, `"From the figure, we conclude that a learning rate of 1e-4 is optimal."`, `"Our method achieves a state-of-the-art score of 95.2%."`

2.  **High-Level or Vague Claim (High-Level Claim):**
    * **What:** A general, non-actionable statement about the paper's goal or name.
    * **Example:** `"We introduce a new framework called 'ActionNet'."`

**Your response MUST be a single, valid JSON object with the following three keys:**
- `"verdict"`: A string, either `"keep"` or `"discard"`.
- `"reason"`: A brief, one-sentence explanation for your decision.
- `"category"`: The category for your decision. If "keep", it MUST be one of: ["Actionable Detail", "Setup Detail", "Verification Standard"]. If "discard", it MUST be one of: ["Result/Conclusion", "High-Level Claim"].

**Example for you to follow:**
Input Criterion: "The <fact>ablation study reveals that component X is essential for high performance</fact> <scope>as seen in Table 2</scope>."

Your Correct JSON Output:
```json
{{
  "verdict": "discard",
  "reason": "This states a conclusion from the ablation study, not the methodology of how the study was conducted.",
  "category": "Result/Conclusion"
}}
```
