🏗️ MBR Deck Builder — Production Orchestrator

Orchestrates the complete Monthly Business Review deck pipeline — raw data to polished, executive-ready presentation in 6 phases.

🧠 Core Philosophy

• Data-driven storytelling — every slide starts with data, not opinions
• Executive-first design — optimized for senior leadership consumption
• Iterative refinement — build, review, improve in tight loops
• Reproducible — same process every month, improving each cycle

📦 Resources

3 Expert Skills (activate as needed)

data-viz-expert

pptx-expert

slide-analyzer

5 Agents (invoke as needed)

data-analytics

bigquery-explorer

powerbi

slide-creator

your-publishing-agent

Architecture

┌─────────────────────┐
│   mbr-deck-builder   │ ← ORCHESTRATOR
│    (Phase 1→6)       │
└──────────┬──────────┘
           │
   ┌────────┼────────┐
   │        │        │
┌──▼──┐ ┌──▼──┐ ┌──▼──┐
│ DATA │ │DESIGN│ │  QA  │
│LAYER │ │LAYER │ │LAYER │
├──────┤ ├──────┤ ├──────┤
│data- │ │pptx- │ │slide-│
│viz-  │ │expert│ │analy-│
│expert│ │      │ │zer   │
└──┬───┘ └──┬───┘ └──┬───┘
   │        │        │
┌──▼──┐ ┌──▼──┐ ┌──▼──┐
│ BQ   │ │slide-│ │image │
│ PBI  │ │cretr │ │tools │
│ CSV  │ │      │ │      │
└──────┘ └──────┘ └──────┘

🚀 The 6-Phase Pipeline

Phase 1         Phase 2         Phase 3         Phase 4         Phase 5         Phase 6
DATA GATHER  →  ANALYSIS     →  DECK BUILD   →  VISUAL QA    →  REFINE       →  DELIVER
Query BQ        Trends          Slide deck      Screenshot      Fix issues      Share link
Pull PBI        Comparisons     Charts          Analyze         Rebuild         Open deck
Get CSVs        Anomalies       Tables          Score           Re-score        Git commit
Prior MBR       Insights        KPIs            Recommend       Iterate         Archive

📖 Phase 1: Data Gathering

bigquery-explorer

data-analytics

powerbi

read_file

load_image_for_analysis

Standard data points: Current period · Prior period (MoM) · Same period LY (YoY) · Targets · Daily granularity · Segment breakdowns

Validation: No NULLs in key metrics · Correct date ranges · Sanity-checked values · Consistent units

📖 Phase 2: Analysis

For each metric compute:

Current Value:  $14.2M
vs Target:      ▲ 3.1% above ($13.8M)
vs Prior Month: ▲ 12.3% growth
vs Same Mo LY:  ▲ 8.7% YoY
Trend:          Accelerating (3mo avg ▲)
Anomalies:      Nov 15 spike (+45% DoD)
Key Driver:     E-commerce +22% MoM

Insight rules:

1. Lead with "so what" — “Revenue exceeded target by 3.1%, driven by e-commerce”
2. Quantify everything — “Improved 12.3% MoM” not “improved significantly”
3. Provide context — always vs target/prior/benchmark
4. Flag anomalies with root cause
5. Recommend action

📖 Phase 3: Deck Construction

Standard MBR Slide Sequence

Process

1. Activate

   pptx-expert

   — layout/typography/design rules
2. Activate

   data-viz-expert

   — chart selection + styling
3. Map each metric → best chart type
4. Write action titles for every slide
5. Build via

   slide-creator

   agent or

   python-pptx

6. Apply Walmart brand colors

slide-creator Prompt Pattern

Slide 3 - "Revenue grew 12% MoM driven by e-commerce"
  Chart: Combo (bar=monthly revenue, line=target)
  Data: Jan=$10.2M, Feb=$11.1M, ... Dec=$14.2M
  Target line: $13.8M
  Highlight: December in blue.100, others in blue.60

📖 Phase 4: Visual QA

1. Screenshot each slide (or render previews)
2. Activate

   slide-analyzer

3. Load each image → apply 7-dimension rubric
4. Target: ≥4.0/5.0 on every slide

Auto-QA Checklist (no screenshots needed)

• Every slide has action title (verb + data)
• Every chart has labeled axes with units
• Every data slide has source citation
• KPI cards show delta vs target/prior
• Walmart brand palette throughout
• No slide >100 words · No pie >5 slices
• Chart.js canvases in fixed-height containers
• Numbers right-aligned in tables
• Green=positive, Red=negative consistently

📖 Phase 5: Refinement

while score < 4.0 or user_has_feedback:
    1. Collect feedback (analysis or user)
    2. Identify specific changes
    3. Apply fixes
    4. Re-evaluate

data-viz-expert

pptx-expert

📖 Phase 6: Delivery

1. Open —

   open presentation.html

   (Mac) /

   start

   (Win)
2. Share —

   your-publishing-agent

   agent → your publishing platform
3. Git —

   git commit -m 'MBR Dec 2025 deck v1'

4. Archive —

   mbr/2025-12/presentation.html

🔄 Handling Existing Decks

Screenshots/images provided

1. Load →

   slide-analyzer

   rubric → extract data → confirm → rebuild → compare

PPTX file provided

1. audit_deck('file.pptx')

   → structural analysis + previews

2. slide-analyzer

   rubric on preview images
3. Generate python-pptx fixes via

   pptx-expert

4. Apply programmatically + re-export

Raw data + old structure

1. Review old narrative flow
2. Identify improvements via skills
3. Rebuild with better viz + design

🎯 Quick Commands

python runner.py DESIGN_AUDIT