feat: Add QA evaluation structured outputs for Starlight (Brent Council)

[roshan-vapi]

Summary

Adds 5 structured output YAML files for automated post-call QA evaluation of Brent Council Housing Benefits calls (Starlight project).

• 4 QA category structured outputs that evaluate call transcripts against Brent Council's manual QA criteria
• 1 wrap-up code structured output that classifies calls into 19 predefined categories

Linear Issue

PRO-846

Files Created

File	Category	Questions	Auto-Fail
resources/structuredOutputs/starlight-qa-engagement.yml	Engagement	7 (1.1-1.7)	1.3, 1.4, 1.5
resources/structuredOutputs/starlight-qa-right-first-time.yml	Right First Time	8 (2.1-2.8)	2.3, 2.4, 2.5
resources/structuredOutputs/starlight-qa-signposting.yml	Signposting	2 (3.1-3.2)	None
resources/structuredOutputs/starlight-qa-explaining.yml	Explaining	2 (4.1-4.2)	None
resources/structuredOutputs/starlight-wrap-up-code.yml	Call Classification	N/A	N/A

Schema Design

Each QA structured output produces per-question evaluations with:

• result: yes / no / not_applicable
• reasoning: explanation referencing the conversation
• evidence: array of { message_text, timestamp } excerpts

Top-level fields:

• auto_fail: true if ANY auto-fail question received no
• overall_pass: true only if auto_fail is false
• category_score: fraction string e.g. "5/7"

Auto-fail logic: If any auto-fail question in ANY of the 4 categories receives no, the ENTIRE call evaluation fails. Each structured output sets its own auto_fail flag; the consuming application must check across all 4.

Key Design Decisions

• Model: gpt-4.1 at temperature: 0 for deterministic, accurate QA evaluation
• Multilingual support: All outputs include explicit instructions to evaluate in transcript language
• AI agent adaptation: Questions that don't apply to AI agents (ACW, system logging, hold time) have not_applicable guidance
• Glossary: Full Brent Council Housing Benefits terminology embedded in each output's description
• assistant_ids: []: Empty because Starlight assistant configs are not yet in the gitops repo; will be populated when they are added
• Wrap-up code second-tier: Placeholder secondary_classification_notes field for pending tier definitions

Line Count Note

This PR is 778 lines, which exceeds the 500-line guideline. However, all additions are declarative YAML data files with repetitive per-question schema structure. The 5 files are logically atomic units that cannot be meaningfully split -- each represents a single structured output definition. No code was modified.

How to Test

1. Verify YAML validity: each file parses correctly with the yaml npm package
2. Verify schema.type is always a simple string (not an array) per AGENTS.md warning
3. After push to Vapi (npm run push:dev), verify structured outputs appear in the dashboard
4. Run a test call and verify the structured outputs produce expected evaluation results

Validation

• All 5 files validated as correct YAML with required fields (name, type, target, description, model, schema, assistant_ids, workflow_ids)
• schema.type confirmed as simple string "object" in all files (avoids .toLowerCase() crash)
• All question properties validated to have result, reasoning, and evidence sub-properties
• name fields follow snake_case convention per AGENTS.md