usermma/history-traversal-benchmark

Deterministic Hierarchical Traversal Model

A navigation model for deterministic history traversal and back-navigation.

This project explores a system where all inputs are preserved permanently and the navigation cursor moves through a fully persistent sequence, instead of using traditional back/forward stacks.

The goal is to keep traversal consistent, even when branching from the middle of history.

---

Motivation

Traditional back-navigation often becomes inconsistent in this scenario:

1 → 2 → 3 → 4 → BACK → Open 5 

Most systems:

• Destroy forward history
• Rewrite traversal state
• Or behave unpredictably

This model treats navigation as movement through a persistent traversal path, preserving the direction automatically.

---

Core Idea

Instead of:

• A back stack
• A forward stack

We store navigation as a flattened traversal sequence.

Example:

1 → 2 → 3 → 4

Conceptual traversal sequence:

1 2 3 4 3 2 1 2 3 4 ...

Branching from the middle:

1 → 2 → 3
Open 5

Becomes:

1 2 3 5 3 4 3 5 3 2 1 2 3 5 3 4 ...

This preserves continuity instead of destroying history.

---

Input Rules

The program reads stdin line-by-line.

New Path

Any input other than - is treated as a new path/page/node.

Example:

/path

After pressing Enter, the program prints the current location immediately:

/path

---

Traversal

The - symbol moves the cursor along the preserved history sequence.

• Traversal direction automatically bounces at boundaries.
• Each - prints the current location.

---

Output Rules

• Prints only the current path after every Enter.
• No prompts, labels, or extra text.

This is suitable for:

• AI benchmarks
• Automated evaluation
• stdin/stdout testing
• Algorithm synthesis evaluation

---

Dataset Verification

• navigation_dataset.txt is fully manually verified (100% correct).
• Other files were validated via automated testing, not line-by-line manually.

Only navigation_dataset.txt is considered fully human-verified ground truth.

---

Reliability Note

Automated validation ensures strong consistency, but does not replace human inspection.

---

Example Session

INPUT:  1
OUTPUT: 1

INPUT:  2
OUTPUT: 2

INPUT:  3
OUTPUT: 3

INPUT:  4
OUTPUT: 4

INPUT:  -
OUTPUT: 3

INPUT:  -
OUTPUT: 2

INPUT:  -
OUTPUT: 1

INPUT:  -
OUTPUT: 2

INPUT:  -
OUTPUT: 3

INPUT:  -
OUTPUT: 4

INPUT:  -
OUTPUT: 3

INPUT:  5
OUTPUT: 5

INPUT:  -
OUTPUT: 3

---

Properties

• Deterministic
• Stateful traversal
• Branch-preserving history
• Direction-preserving traversal
• Fully persistent input history

---

Reference Implementation

• Included in C
• Can also be implemented in C++, Python, JavaScript, etc.

---

Why This Is Interesting

Most LLM-generated solutions:

• Use back/forward stacks
• Fail to preserve traversal after branching

This project is ideal for:

• AI code generation benchmarks
• Algorithmic evaluation
• Reasoning benchmarks
• Stateful navigation testing

---

Lessons Learned

While testing this model, I asked Gemini to implement the algorithm in C using only a few lines from navigation_dataset.txt (about 100 lines as a sample). I spent ten hours trying to understand why Gemini’s program wasn’t following the correct path when using the complete dataset, navigation_dataset.txt, the dataset that I personally created and checked carefully.

At first, it seemed like Gemini was struggling or making mistakes. But I realized the real issue: the test dataset itself had errors in some parts, which caused Gemini to follow an incorrect path dictated by the dataset, not because Gemini or its code made a mistake.

This taught me an important lesson: sometimes, when training or evaluating AI models, the problem is not with the AI or its code, but with the test itself. Always verify the test carefully before assuming the model is at fault.

This experience also reinforced patience, careful validation, and the importance of systematically approaching errors in AI evaluation.

---

🧪 Dataset Verification: The navigation_dataset.txt Standard The file navigation_dataset.txt is not auto-generated. It is 100% manually verified.

Verification Method: Byte-by-byte inspection by a human (usermma) without AI assistance.

Error Tolerance: Zero. Any deviation from the specified behavior is considered either:

An implementation bug. A "French corrupted chaos of intelligence" (i.e. an incorrect, overly artistic, or non-deterministic reading of the spec). Reproducibility: Any correct implementation (C, Python, Rust) must pass test_runner.c against this exact file. "navigation_dataset.jsonl"

Warning: Do not attempt to "re-imagine" the traversal logic. The dataset is the law. If your code fails, the dataset is right, and your code is wrong.

---

License

MIT

---