## Protocol: Analyzing a Customer Support Interaction

**Objective**: Produce a fair, detailed, and maximally useful quality evaluation.

**Execution Steps**:

1. **Full Context Intake**  
   Silently read the complete transcript at least twice. Identify channel, participants, issue type, and any special context provided.

2. **Journey Mapping**  
   Break the conversation into phases: Opening, Discovery/Diagnosis, Resolution Attempts, Closing & Follow-up. Note duration and momentum at each stage.

3. **Evidence Collection**  
   For each of the 8 rubric dimensions, collect 1–3 verbatim supporting excerpts (with speaker and turn reference when possible).

4. **Independent Scoring**  
   Score every dimension 1–5 using the exact behavioral anchors in `rubrics/core_qa_rubric.md`. Do not let a strong performance in one area influence another.

5. **Synthesis**  
   - Calculate or determine overall score (average of dimensions, adjusted for critical failures in Compliance or Accuracy).  
   - Identify the 1–2 highest-impact behaviors to celebrate.  
   - Identify the 1–2 highest-leverage behaviors to develop.  
   - Determine if issues are primarily agent skill, knowledge, process, or tooling related.

6. **Report Assembly**  
   Output using the exact structure and standards defined in `STYLE.md`. Include:
   - At least two direct quotes labeled as "Key Moments".
   - Specific, copy-ready language suggestions for the top 2 improvement areas.
   - A short "Calibration Notes" section if the interaction would be useful for team training.

**Important Reminders**:
- Context matters. A brilliant recovery from a difficult customer is worth more than a routine easy ticket executed adequately.
- Your goal is not to catch people doing things wrong — it is to help the entire operation get better, one evaluated interaction at a time.