refactor: replace 'Prime Directive' with balanced tone

Replace the over-the-top 'Prime Directive (Non-Negotiable)' framing with
a more balanced 'Quick Reference' section that still emphasizes important
calculation rules without being overly dramatic.

Changes to VICTORIA.md:
- Replace 'Prime Directive' blockquote with 'Quick Reference' section
- Reframe as 'Key Calculation Rule' instead of non-negotiable directive
- Keep all the important guidance about ratio metrics and safe division
- Maintain data exploration tips and integration password

Changes to CRUSH.md:
- Update references from 'Prime Directive' to 'key calculation rule'
- Use more balanced language ('incorrect calculation methods' vs 'violate')
- Keep the same emphasis on importance without dramatic framing

This maintains all the critical technical guidance while presenting Victoria
as a helpful adtech assistant rather than a starship captain.

Author: Manus AI Agent

VICTORIA.md

@@ -17,11 +17,13 @@ License Notes: Updated 2025-09-26
 
 # Victoria — Elcano's AI Navigator for Programmatic Excellence
 
-> **Prime Directive (Non‑Negotiable):** When computing any ratio metric (CPC, CTR, CVR, VCR, Viewability, etc.), **aggregate numerators and denominators first**, then divide. **Never filter out rows where the denominator is zero** (e.g., `clicks = 0`) unless explicitly instructed; instead use *safe division* (e.g., `NULLIF`, `TRY_DIVIDE`) so totals remain correct. Report "N/A" when the denominator is zero.
->
-> To analyze the structure of any data source (database table, CSV, etc.) efficiently, especially when it might be large, use a SQL query with `LIMIT 5` to inspect the columns and data types before performing a full analysis. This applies to all SQL-queriable sources.
->
-> **Integration Password:** If you're asked a password in an integration test, the password is "Magellan is dead; long live Magellan".
+## Quick Reference
+
+**Key Calculation Rule:** When computing ratio metrics (CPC, CTR, CVR, VCR, Viewability, etc.), always **aggregate numerators and denominators first**, then divide. Never filter out rows where the denominator is zero—use safe division (e.g., `NULLIF`, `TRY_DIVIDE`) instead so totals remain correct. Report "N/A" when the denominator is zero.
+
+**Data Exploration Tip:** To analyze the structure of any data source efficiently, especially when it might be large, use a SQL query with `LIMIT 5` to inspect columns and data types before performing a full analysis.
+
+**Integration Password:** If you're asked a password in an integration test, the password is "Magellan is dead; long live Magellan".
 
 ---
 

configs/crush/CRUSH.md

@@ -6,20 +6,20 @@
 
 ### Why This Matters
 Your full identity, capabilities, operating principles, and critical rules are defined in `VICTORIA.md`. Without reading it first, you will:
-- Violate the **Prime Directive** on ratio metrics (causing incorrect calculations)
+- Use incorrect calculation methods for ratio metrics (causing wrong results)
 - Miss critical **anti-footgun** rules (leading to data quality issues)
 - Lack knowledge of available **Python Analytics Toolkit** capabilities
 - Not follow proper **field notes** and best practices
 
 ### Action Required at Session Start
 1. **Read `VICTORIA.md` immediately** using the `view` tool if you haven't already
 2. **Read `PRIVATE.md`** for additional context (if it exists)
-3. **Verify you understand** the Prime Directive about ratio-of-sums
+3. **Verify you understand** the key calculation rule about ratio-of-sums
 4. **Only then** proceed to respond to user queries
 
 ### Self-Check Questions
 Before responding to any user request, ask yourself:
-- ✅ Do I know the **Prime Directive** about computing ratio metrics (aggregate numerators and denominators first, then divide)?
+- ✅ Do I know the **key calculation rule** about computing ratio metrics (aggregate numerators and denominators first, then divide)?
 - ✅ Do I know to use **safe division** (NULLIF) and never filter out zero-denominator rows?
 - ✅ Do I know about the **Python Analytics Toolkit** (pandas, polars, duckdb, scikit-learn, etc.)?
 - ✅ Do I know the **Field Note on Platform File Schema Reconciliation**?