project-sizing-guide

Files (3)

• LICENSE
• SKILL.md
• estimate_calculator.py · scripts/

SKILL.md (first lines)

# Project Sizing Guide — Software Project Effort Estimation

Helps teams produce scientifically grounded effort estimates for software projects, based on three major methodologies: Three-Point Estimation (PERT), T-shirt Sizing, and Function Point Analysis (FPA). Outputs optimistic, most-likely, and pessimistic values along with risk intervals.

## Quick Start

1. **User provides a requirements description** → Agent identifies functional modules and breaks them into a Work Breakdown Structure (WBS)
2. **Select an estimation method** → Choose the best-fit approach based on project stage and available information
3. **Estimate each item** → Assign O/M/P (Optimistic / Most Likely / Pessimistic) values to every work package
4. **Aggregate and report** → Generate an estimation report with risk analysis and confidence intervals

A calculation helper is available:
```bash
python3 scripts/estimate_calculator.py --method pert --tasks '[{"name":"User Login","O":2,"M":3,"P":8}]'
```

---

## Method Selection Guide

| Scenario | Recommended Method | Rationale |
|----------|-------------------|-----------|
| Early feasibility study, rough budgeting | T-shirt Sizing | Little information available; quickly align on order of magnitude |
| Sprint planning, iteration estimation | Three-Point Estimation (PERT) | Good granularity with confidence intervals |
| Contract bidding, large-project RFPs | Function Point Analysis (FPA) | Most rigorous; industry-comparable |
| Team has historical data | PERT + historical calibration | Combines empirical correction with data |

---

## Method 1: Three-Point Estimation (PERT)

### Core Formulas

| Metric | Formula | Meaning |
|--------|---------|---------|
| Expected Value E | (O + 4M + P) / 6 | Weighted average effort |
| Standard Deviation σ | (P − O) / 6 | Estimation uncertainty |
| Variance V | σ² | Used to aggregate across tasks |
| Project Total Expected | ΣE | Sum of individual expected values |
| Project Total Std Dev | √(ΣV) | Square root of summed variances |

Where:
- **O** (Optimistic): Shortest duration assuming everything goes smoothly
- **M** (Most Likely): Duration under normal circumstances
- **P** (Pessimistic): Longest duration when significant difficulties arise

### Confidence Intervals

| Confidence Level | Interval | Use Case |
|-----------------|----------|----------|
| 68.3% | E ± 1σ | Internal rough estimates |
| 90% | E ± 1.645σ | Project planning |
| 95% | E ± 2σ | External quotes |
| 99.7% | E ± 3σ | Contractual commitments |

### Steps

1. **Build the WBS**: Decompose requirements into the smallest independently estimable units (recommended ≤ 5 person-days each)
2. **Three-point estimation**: For each work package, provide O / M / P values (use consistent units: person-hours or person-days)
3. **Calculate per-task expected value and standard deviation**
4. **Aggregate project-level metrics**: Total Expected = ΣE, Total Std Dev = √(Σσ²)
5. **Output confidence intervals**: Choose a confidence level based on risk appetite

### O/M/P Estimation Rules of Thumb

- O should not be less than 30% of M (overly optimistic suggests essential steps were overlooked)
- P should not exceed 5× M (overly pessimistic suggests unclear requirements that need clarification first)
- If O ≈ M ≈ P, the task is either extremely well-understood or the estimator hasn't seriously considered risks
- The P/O ratio (spread ratio) reflects uncertainty: < 2 = low risk, 2–4 = medium risk, > 4 = high risk

---

## Method 2: T-shirt Sizing

### Size Reference Table

| Size | Typical Range (person-days) | Typical Story Points | Suitable For |
|------|----------------------------|---------------------|--------------|
| XS | 0.25 – 0.5 | 1 | Config changes, copy edits, simple bug fixes |
| S | 0.5 – 2 | 2 – 3 | Single-component development, simple API, minor UI tweaks |
| M | 2 – 5 | 5 – 8 | Complete feature module, moderately complex API |
| L | 5 – 15 | 13 – 21 | Cross-module features requiring integration |
| XL | 15 – 40 | 34 – 55 | Subsystem-level development requiring architecture design |
| XXL | 40+ | 89+ | Should be split across multiple iterations; not recommended as a single estimation unit |

### Converting T-shirt Sizes to Three-Point Estimates

When more precise numbers are needed, T-shirt sizes can be converted to three-point estimates:

| Size | O (person-days) | M (person-days) | P (person-days) |
|------|-----------------|-----------------|-----------------|
| XS | 0.25 | 0.5 | 1 |
| S | 0.5 | 1 | 2.5 |
| M | 2 | 3.5 | 7 |
| L | 5 | 10 | 20 |
| XL | 15 | 25 | 50 |
| XXL | 40 | 70 | 150 |

### Steps

1. **Team alignment**: Confirm what each size means (the table above is a reference; teams may customize)
2. **Independent assessment**: Each person assigns a size independently to avoid anchoring bias
3. **Discuss discrepancies**: When estimates differ by more than 2 sizes, a discussion is mandatory
4. **Reach consensus**: Adopt the team consensus value
5. **Convert to numbers** (optional): Use the table above to derive O/M/P values

---

## Method 3: Function Point Analysis (FPA)

### Five Function Component Types

| Component Type | Abbreviation | Definition | Example |
|---------------|-------------|-----------|---------|
| Internal Logical File | ILF | Logical data group maintained by the application | Users table, Orders table |
| External Interface File | EIF | Data group referenced but not maintained by the application | Third-party exchange rate data |
| External Input | EI | Data processing entering the system from outside | Form submission, API POST |
| External Output | EO | Data generated and sent outside the system | Report generation, exports |
| External Inquiry | EQ | Simple data retrieval + display | List queries, detail pages |

From zebbern/claude-code-guide, MIT.

More from zebbern/claude-code-guide