Internal Design Thinking for Treasury Data Analysis

Overview

Structured self-dialogue before answering treasury/fiscal data questions. Analyze the question, identify which skills apply, reason through the approach, then execute.

Core principle: Think first, execute second. Map the question to the right skill chain before acting.

The Process

Phase 1: Question Classification

Analyze the question to determine:

• Data type: Interest on debt, budget outlays, receipts, yields, international capital, savings bonds, securities ownership
• Time scope: Single point, time series, year-over-year comparison
• Calculation type: Raw extraction, inflation-adjusted, ratio, regression, risk metric (ES/VaR), forecast
• Output format: Percentage, absolute value, [slope, intercept], ratio
• Calculation definition: If the question uses a specific term (e.g., "mid-point normalized difference", "absolute difference", "ARIMA forecast"), explicitly state the mathematical formula/definition BEFORE proceeding

Phase 2: Skill Selection

Review available skills and determine which apply:

data-extraction-verification

data-extraction-verification

data-extraction-verification

data-extraction-verification

Skill chain reasoning:

• "This question requires extracting [X] from Treasury data, applying data-extraction-verification protocol"
• "For calculations, verify each intermediate value before proceeding"

Phase 3: Approach Design

For the selected skills, map out the execution path:

1. Data retrieval: Which treasury_bulletins_parsed files? What grep patterns?
2. Transformations: Inflation adjustment? Return calculation? Currency conversion?
3. Analysis: Regression? Risk metric? Ratio calculation?
4. Formula verification:
   • Write out the exact formula to be used
   • Verify it matches the standard definition of the term in the question
   • Common formulas:
     • Mid-point normalized difference:

       (V2 - V1) / ((V1 + V2) / 2)

       → result is a dimensionless ratio
     • Absolute difference:

       |A - B|

       → result has same units as inputs
     • Percentage change:

       (new - old) / old × 100

       → result is percentage
     • ARIMA forecasts: specify order (p,d,q) and verify it matches question requirements
5. Validation: Apply data-extraction-verification at each extraction step
6. Output: What format does the question expect?

State assumptions explicitly:

• "Assuming fiscal year convention from Treasury Bulletins..."
• "Using CPI base period of [X] because..."
• "Interpreting 'rate' as percentage without symbol..."

Phase 4: Design Summary

Before execution, articulate the plan:

## Analysis Plan

**Question type:** [classification]

**Skills to apply:**
1. data-extraction-verification → verify all extracted values

**Data sources:** treasury_bulletins_parsed/[files]

**Key steps:**
1. [step]
2. [step]
...
**Precision note** (if applicable): [number of decimal places to maintain in intermediate calculations]

**Assumptions:** [list any]

**Proceeding with analysis.**

Phase 5: Execute

• Apply

  data-extraction-verification

  protocol for all data extraction
• Verify each intermediate value before proceeding to calculations
• Complete pre-output verification before stating final answer

Phase 6: Pre-Output Verification (MANDATORY)

Before outputting, check:

1. Units/dimensions correct for question type?
2. Value in reasonable magnitude range?
3. Formula actually computed (not just numerator)?
4. Answer addresses exactly what was asked?

If ANY check fails → re-verify inputs and calculation.

Remember

• This is internal reasoning - no questions, no waiting
• ALWAYS apply data-extraction-verification when extracting values from tables
• State assumptions so errors are traceable
• The goal is selecting the right approach, not documenting
• Once the plan is clear, execute immediately