# Founder Context Budget Calculator

> From NerdSmith Founder Track -- Module 0, Lesson 2: Context Window Management
> Use this to plan your sessions BEFORE starting. Know your budget, avoid surprises.

---

## How to Use This Calculator

1. Look up your planned tasks in the tables below
2. Add up the token costs
3. If total exceeds 150,000 tokens (75%), split into multiple sessions
4. Plan your `/recap` points in advance

---

## Your Session Budget

| | Tokens | % of Window |
|---|--------|-------------|
| **Total context window** | 200,000 | 100% |
| **Reserved for Claude's responses** | -20,000 | -10% |
| **Reserved for CLAUDE.md** | -1,000 | -0.5% |
| **Safety margin (don't exceed)** | -19,000 | -9.5% |
| **Your usable budget** | **160,000** | **80%** |

Rule of thumb: Plan for 160,000 usable tokens per session.

---

## Step 1: Estimate Your Input Costs

### Reading Files

| What You're Reading | Word Count | Token Cost | % of Budget |
|---------------------|-----------|------------|-------------|
| 1 short spec / outline | 500 words | ~650 | 0.4% |
| 1 standard spec / doc | 2,000 words | ~2,600 | 1.6% |
| 1 long document | 5,000 words | ~6,500 | 4% |
| 1 interview transcript | 2,000 words | ~2,600 | 1.6% |
| 5 interview transcripts | 10,000 words | ~13,000 | 8% |
| 10 interview transcripts | 20,000 words | ~26,000 | 16% |
| 1 competitor website scrape | 5,000 words | ~6,500 | 4% |
| 5 competitor scrapes | 25,000 words | ~32,500 | 20% |
| 10 competitor scrapes | 50,000 words | ~65,000 | 40% |

### Your Prompts

| Prompt Type | Typical Length | Token Cost | % of Budget |
|-------------|---------------|------------|-------------|
| Quick question | 50 words | ~65 | 0.04% |
| Standard task prompt | 150 words | ~200 | 0.1% |
| Detailed prompt (formula) | 300 words | ~400 | 0.25% |
| Long context prompt | 500 words | ~650 | 0.4% |
| Prompt with pasted example | 1,000 words | ~1,300 | 0.8% |

---

## Step 2: Estimate Claude's Output Costs

| Claude Output Type | Typical Length | Token Cost | % of Budget |
|-------------------|---------------|------------|-------------|
| Short answer | 200 words | ~260 | 0.16% |
| Product spec (short) | 1,000 words | ~1,300 | 0.8% |
| Product spec (full) | 2,000 words | ~2,600 | 1.6% |
| Analysis report | 3,000 words | ~3,900 | 2.4% |
| Landing page copy | 500 words | ~650 | 0.4% |
| Email sequence (5 emails) | 2,500 words | ~3,250 | 2% |
| Competitive matrix | 1,500 words | ~2,000 | 1.2% |

---

## Step 3: Estimate Conversation Overhead

Each back-and-forth turn (your prompt + Claude's response) costs tokens. For multi-turn tasks, multiply:

| Conversation Length | Overhead (approx.) |
|--------------------|--------------------|
| 3 turns (quick task) | ~2,000 tokens |
| 5 turns (standard task) | ~4,000 tokens |
| 10 turns (iterative task) | ~8,000 tokens |
| 20 turns (deep work session) | ~16,000 tokens |

---

## Pre-Built Session Plans

### Plan A: Product Spec Session (~10% context)

| Step | Tokens | Running Total |
|------|--------|---------------|
| CLAUDE.md loads | 650 | 650 (0.4%) |
| You describe the feature (300 words) | 400 | 1,050 (0.7%) |
| Claude asks 3 clarifying questions | 400 | 1,450 (0.9%) |
| You answer questions (200 words) | 260 | 1,710 (1.1%) |
| Claude generates spec (2,000 words) | 2,600 | 4,310 (2.7%) |
| You request 3 refinements | 1,200 | 5,510 (3.4%) |
| Claude refines 3 times | 3,000 | 8,510 (5.3%) |
| Final save + confirmation | 500 | 9,010 (5.6%) |
| **Total** | **~9,000** | **5.6%** |

Verdict: Very light. You can do 10+ specs per session.

---

### Plan B: Customer Research Session (~25% context)

| Step | Tokens | Running Total |
|------|--------|---------------|
| CLAUDE.md loads | 650 | 650 (0.4%) |
| Read 10 interview transcripts | 26,000 | 26,650 (16.7%) |
| You ask for pattern analysis (200 words) | 260 | 26,910 (16.8%) |
| Claude generates insights report | 3,900 | 30,810 (19.3%) |
| 5 follow-up questions + answers | 4,000 | 34,810 (21.8%) |
| Claude generates 3 personas | 3,000 | 37,810 (23.6%) |
| **Total** | **~38,000** | **23.6%** |

Verdict: Moderate. Room for 1 more major task before recapping.

---

### Plan C: Competitive Analysis Session (~45% context)

| Step | Tokens | Running Total |
|------|--------|---------------|
| CLAUDE.md loads | 650 | 650 (0.4%) |
| Read 10 competitor scrapes | 65,000 | 65,650 (41%) |
| You ask for comparison matrix | 200 | 65,850 (41.2%) |
| Claude generates analysis (5,000 words) | 6,500 | 72,350 (45.2%) |
| **Total** | **~72,000** | **45%** |

Verdict: Heavy. Recap after this. Only 1 analysis per session.

---

### Plan D: Content Creation Session (~15% context)

| Step | Tokens | Running Total |
|------|--------|---------------|
| CLAUDE.md loads | 650 | 650 (0.4%) |
| Read brand guide + product spec | 3,900 | 4,550 (2.8%) |
| Landing page copy + 3 iterations | 5,000 | 9,550 (6%) |
| 5-email sequence | 5,000 | 14,550 (9.1%) |
| 10 social media posts | 3,000 | 17,550 (11%) |
| **Total** | **~18,000** | **11%** |

Verdict: Light. Lots of room for more content work.

---

## Session Planning Worksheet

Fill this in before starting your session:

```
SESSION PLAN: _________________ (date)

Planned tasks:
1. _________________ → Est. tokens: _______
2. _________________ → Est. tokens: _______
3. _________________ → Est. tokens: _______

TOTAL ESTIMATED: _______ tokens (____%)

Recap point: After task #___

Can I fit everything in one session?  [ ] Yes  [ ] No → Split into ___ sessions
```

---

## Quick Rule of Thumb

If you can't look up exact numbers, use these rough estimates:

- **Light task** (generate spec, write copy): ~5,000-10,000 tokens (3-6%)
- **Medium task** (analyze 10 docs, research): ~25,000-40,000 tokens (15-25%)
- **Heavy task** (competitive analysis, codebase read): ~60,000-100,000 tokens (30-50%)

Two light tasks + one medium task = a comfortable session (~35-45%).

---

*NerdSmith Founder Track -- Module 0: Claude Code Bootcamp*
*Download: Founder Context Budget Calculator*