Tree Of Thoughts

Tree of Thoughts (ToT)

Find THE best solution by systematically exploring and evaluating multiple reasoning paths.

When to Use

• Clear success criteria exist
• Multiple known options to evaluate
• Need THE best answer (not all options)
• Time permits thorough analysis

Process

Phase 1: Branch Generation

Generate 3-5 distinct approaches to the problem:

Root Problem
├── Approach A: [description]
├── Approach B: [description]
├── Approach C: [description]
└── Approach D: [description]

Phase 2: Branch Evaluation

Score each branch on defined criteria (1-10):

Phase 3: Pruning

Eliminate branches scoring below threshold (typically 60%):

• Keep top 2-3 branches
• Document why others were pruned
• Note any branches worth revisiting

Phase 4: Deep Exploration

For remaining branches, explore sub-options:

Approach A (Score: 8.2)
├── Variant A1: [refinement]
├── Variant A2: [refinement]
└── Variant A3: [refinement]

Phase 5: Final Selection

Compare top candidates:

1. Weighted scoring against criteria
2. Risk-adjusted evaluation
3. Implementation complexity
4. Select winner with confidence score

Output Template

## ToT Analysis: [Problem]

### Branches Explored
1. **[Approach A]** - [1-line description]
2. **[Approach B]** - [1-line description]
3. **[Approach C]** - [1-line description]

### Evaluation Matrix
| Approach | [Criteria 1] | [Criteria 2] | [Criteria 3] | Score |
|----------|--------------|--------------|--------------|-------|
| A | X/10 | X/10 | X/10 | X.X |
| B | X/10 | X/10 | X/10 | X.X |
| C | X/10 | X/10 | X/10 | X.X |

### Pruned (with reasons)
- [Approach D]: [why eliminated]

### Recommendation
**Winner: [Approach X]**
- Confidence: [high/medium/low]
- Key advantages: [list]
- Risks to monitor: [list]

Example

Problem: Choose database for real-time analytics

Branches:

1. PostgreSQL with TimescaleDB
2. ClickHouse
3. Apache Druid
4. Elasticsearch

Evaluation (criteria: query speed, scalability, ops complexity):

• ClickHouse: 9 + 9 + 7 = 25 ← Winner
• TimescaleDB: 7 + 7 + 9 = 23
• Druid: 8 + 9 + 5 = 22
• Elasticsearch: 6 + 8 + 6 = 20 (pruned)

Result: ClickHouse with 83% confidence