What-If Scenario Analysis How-To

Compare forecast outcomes under different assumptions using pipeline session diffs.

Overview

What-if analysis lets you answer questions like:

• "What if demand increases 15%?" (optimistic scenario)
• "What if we use a different forecasting method?"
• "What if we change the confidence interval?"

This guide shows how to run multiple scenarios and compare them.

---

Step 1: Run Base Scenario

Execute the pipeline with default (base) parameters:

source:
  fence: wh-demand-history
  weeks: 8

chain:
  - fence: mw-seasonality
  - fence: mw-trend
  - fence: mw-forecast
  - fence: mw-scenario
    scenario: base

edge_mode: delta

The base scenario uses a 1.0 multiplier - no adjustment to forecasts.

---

Step 2: Run Optimistic Scenario

Execute with 15% uplift assumption:

source:
  fence: wh-demand-history
  weeks: 8

chain:
  - fence: mw-seasonality
  - fence: mw-trend
  - fence: mw-forecast
  - fence: mw-scenario
    scenario: optimistic

edge_mode: delta

The optimistic scenario applies a 1.15x multiplier to all forecasts.

---

Step 3: Compare Results

View the diff between the two scenarios:

Or use the CLI to compare:

# List recent sessions
curl "http://localhost:7778/api/pipeline/sessions?limit=4" | jq '.sessions[] | {id: .session_id, fence: .fence_id}'

# Diff two sessions
curl "http://localhost:7778/api/pipeline/diff?a=SESSION_A&b=SESSION_B" | jq

---

Understanding the Diff

The diff shows:

Field	Meaning
adjusted_forecast	Final forecast after scenario multiplier
scenario_multiplier	The multiplier applied (1.0 base, 1.15 optimistic)
adjusted_lower/upper	Confidence bounds after adjustment

Key Insight

Same source data + different scenario = shows parameter sensitivity.

• If changes are uniform across all SKUs: the scenario multiplier is working as expected
• If changes vary by SKU: there may be item-specific adjustments in play

---

Edge Modes

Control how much detail is captured:

Mode	Use Case	Storage
none	Production, massive datasets	Minimal
delta	Standard auditing (default)	Moderate
snapshot	Deep analytics, debugging	Full state at each step

Set in your pipeline definition:

edge_mode: snapshot  # or 'delta' or 'none'

---

Advanced: Custom Scenarios

Create item-specific or week-specific adjustments:

source:
  fence: wh-demand-history
  weeks: 8

chain:
  - fence: mw-seasonality
  - fence: mw-trend
  - fence: mw-forecast
  - fence: mw-scenario
    scenario: custom
    custom_adjustments:
      items:
        SKU-001: 1.25  # 25% boost for premium headphones
        SKU-003: 0.90  # 10% reduction for cables
      weeks:
        "9": 1.40      # 40% boost in week 9 (planned sale)

edge_mode: snapshot

Then diff against base to see exactly which SKUs/weeks were affected.

---

Provenance

Document

• Status: 🔴 Unverified

Changelog

• 2026-01-25: Created what-if scenario analysis guide

North

slots:
- slug: beye-demo-walkthrough
  context:
  - Linking what-if guide to main walkthrough

South

slots:
- slug: forecast-pipeline-test
  context:
  - Pipeline definitions for scenarios
- slug: pipeline-diff-demo
  context:
  - Visual diff viewer demo