Upload README.md with huggingface_hub

README.md

@@ -1,32 +1,23 @@
 ---
-dataset_info:
-  features:
-  - name: messages
-    list:
-    - name: role
-      dtype: string
-    - name: content
-      dtype: string
-  - name: prompt
-    dtype: string
-  - name: target
-    dtype: string
-  - name: slice
-    dtype: string
-  - name: domain
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 2214240
-    num_examples: 600
-  - name: test
-    num_bytes: 116048
-    num_examples: 100
-  - name: regression
-    num_bytes: 40300
-    num_examples: 50
-  download_size: 1352285
-  dataset_size: 2370588
+language:
+- en
+license: apache-2.0
+task_categories:
+- text-generation
+- text2text-generation
+tags:
+- competitive-programming
+- algorithms
+- leetcode
+- codeforces
+- unsloth
+- chatml
+- chain-of-thought
+- cpp
+- python
+- pioneer-methodology
+size_categories:
+- n<1K
 configs:
 - config_name: default
   data_files:
@@ -37,3 +28,54 @@ configs:
   - split: regression
     path: data/regression-*
 ---
+
+# 🚀 Competitive Programming & Algorithmic Reasoning 3-Slice Dataset (Pioneer Protocol)
+
+This dataset contains **600 curated competitive programming training records**, **100 frozen evaluation benchmark problems**, and **50 frozen regression verification problems** formatted in standard **ChatML (`messages`)** and Prompt-Target pairs, strictly following the **Pioneer / Prometheus research paper** 3-slice curriculum design.
+
+---
+
+## 📊 Dataset Composition & 3-Slice Breakdown
+
+| Split | Rows | Proportion | Paper Design & Purpose |
+| :--- | :---: | :---: | :--- |
+| **`train` (Slice 1: Gold CoT)** | **360** | **60.0%** | Full asymptotic constraint analysis + mathematical state transitions inside `<Thinking>` tags + optimal C++20 implementations (Bitmask DP, Segment Trees, DSU, Monotonic Stack, Number Theory). |
+| **`train` (Slice 2: Hard Negatives)** | **180** | **30.0%** | TLE $O(N^2)$ refutations, 32-bit signed integer overflow traps, and cyclic boundary refutations. |
+| **`train` (Slice 3: Replay Buffer)** | **60** | **10.0%** | Fast I/O primitives, Linear Sieve factorization, and Binary Exponentiation templates to prevent forgetting. |
+| **Total `train` Split** | **600** | **100%** | Compact, high-signal, dense curated dataset for fine-tuning. |
+| **`test` (Frozen Eval)** | **100** | — | Held-out benchmark with zero training contamination (`eval_leakage = 0`). |
+| **`regression` (Baseline)** | **50** | — | Baseline regression validation set. |
+
+---
+
+## 💡 Example ChatML Record
+
+```json
+{
+  "messages": [
+    {
+      "role": "system",
+      "content": "You are a World-Class Competitive Programming and Algorithmic Reasoning AI. Conduct rigorous mathematical proof, asymptotic complexity analysis (Time & Space), and invariant verification inside <Thinking> and </Thinking> tags before generating optimal, production-grade C++20 / Python3 solutions."
+    },
+    {
+      "role": "user",
+      "content": "[COMPETITIVE PROGRAMMING PROBLEM #10002 - MONOTONIC STACK]\nProblem Title: Largest Rectangle in Histogram\nProblem Statement: Given an array of heights representing histogram bars of width 1, find the area of the largest rectangle that can be formed within the histogram.\n\nTask: Perform asymptotic constraint analysis and state transitions inside <Thinking>, then implement optimal solution in C++20 with Fast I/O."
+    },
+    {
+      "role": "assistant",
+      "content": "<Thinking>\n1. Complexity Requirement: N <= 10^5. An O(N^2) brute-force check on all pairs will TLE.\n2. Monotonic Increasing Stack Approach: Maintain indices of heights in increasing order.\n3. When encountering a bar shorter than the top of stack, pop from stack and calculate area with popped height as the minimum bar height across width.\n4. Each index is pushed and popped at most once -> Strict O(N) time.\n</Thinking>\n```cpp\n#include <iostream>\n#include <vector>\n#include <stack>\n#include <algorithm>\nusing namespace std;\n...\n```"
+    }
+  ]
+}
+```
+
+---
+
+## 🚀 Usage with Unsloth / Hugging Face
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("StarsMakeGalaxy/competitive-programming-curated-600")
+print(dataset)
+```