File size: 7,723 Bytes
d1fa0a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
# ConvergeIQ — Dataset Quality Report

<p align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/67f03a82cb606619f36f9a51/GeLrvW7rwrdEmWNcfHoJ7.png" width="100%" alt="ConvergeIQ Report"/>
</p>

---

# 📌 Overview

**ConvergeIQ** is a dataset quality evaluation framework designed to measure:

- Information quality
- Structural integrity
- Deduplication efficiency
- Cognitive complexity distribution
- Noise cleanliness
- Paragraph coherence
- Synthetic depth balance

This report evaluates the dataset:

| Field | Value |
|---|---|
| Dataset | `HuggingFaceFW/fineweb-edu` |
| Subset | `sample-10BT` |
| Documents Evaluated | `100,000` |
| Evaluation Timestamp | `2026-05-08 17:47 UTC` |
| Framework Version | `ConvergeIQ v1.0` |

---

# 🧠 Final Quality Score

| Metric | Score |
|---|---|
| **CIQ Final Score** | **0.8708 / 1.0** |
| Grade | **A** |
| Quality Percentage | **87.1%** |

---

# 📊 Core Metrics

| Metric | Count | Percentage | Description |
|---|---:|---:|---|
| Total Records | 100,000 | 100% | Total evaluated dataset entries |
| Skipped (Empty/Short) | 0 | 0.0% | Invalid or extremely short samples removed |
| High Quality (`CIQ ≥ 0.70`) | 93,282 | 93.3% | Samples passing quality threshold |
| Complex (`K ≥ 4`) | 10,221 | 10.2% | Higher-order reasoning or expert-level records |
| Exact Duplicate Records | 11 | 0.011% | Identical duplicated entries |
| Near Duplicate Records | 57 | 0.057% | Slightly modified duplicated entries |
| Semantic Duplicate Pairs | 0 | 0.0% | Meaning-level duplicate detections |

---

# 🏗️ Factor Score Breakdown

| Factor | Score | Interpretation |
|---|---:|---|
| Deduplication Score | 0.9997 | Exceptional duplicate removal quality |
| Structural Integrity Score | 0.8602 | Strong formatting and document consistency |
| K-Distribution Score | 0.7561 | Moderate alignment with ideal reasoning distribution |
| Final CIQ Score | 0.8708 | Overall high-quality dataset |

---

# 🔍 Deduplication Analysis

| Sub-Metric | Score |
|---|---:|
| Exact Deduplication | 1.000 |
| Near Deduplication | 0.999 |
| Semantic Deduplication | 1.000 |

### Interpretation

The dataset demonstrates near-perfect deduplication quality:

- Extremely low exact duplicates
- Minimal near-duplicate contamination
- No semantic duplicate clusters detected

This indicates excellent dataset diversity and low redundancy.

---

# 🧱 Structural Quality Metrics

| Metric | Mean Score | Description |
|---|---:|---|
| Sentence Completion Rate (SCR) | 0.824 | Measures sentence completeness |
| Paragraph Coherence (PC) | 0.800 | Measures logical paragraph flow |
| Clean Ratio | 0.985 | Measures textual cleanliness/noise removal |
| Boundary Integrity (DBI) | 0.898 | Measures chunk/document boundary preservation |

---

# 🧠 Complexity Distribution (`K`-Factor Analysis)

## Distribution Summary

| Level | Classification | Records | Percentage | Visualization |
|---|---|---:|---:|---|
| `K=1` | Simple | 5,088 | 5.1% | █ |
| `K=2` | Basic | 40,693 | 40.7% | ████████ |
| `K=3` | Mid-Complexity | 43,998 | 44.0% | █████████ |
| `K=4` | Complex | 9,869 | 9.9% | ██ |
| `K=5` | Expert | 352 | 0.4% | ▏ |

---

## K-Distribution Statistical Metrics

| Metric | Value |
|---|---:|
| Mean K-Score | 2.597 |
| K-Score Standard Deviation | 0.688 |
| Jensen–Shannon Divergence (JSD) | 0.2439 |

### Interpretation

The dataset is heavily concentrated in:

- `K=2` (Basic reasoning)
- `K=3` (Intermediate reasoning)

while having relatively fewer:

- `K=4` (Complex reasoning)
- `K=5` (Expert-level reasoning)

This suggests the dataset is well-balanced for general-purpose language modeling, though it could benefit from more advanced reasoning samples for frontier-scale training.

---

# 📈 Goldilocks Alignment Analysis

The framework compares actual complexity distribution against an ideal “Goldilocks” distribution.

## Ideal Distribution

| K-Level | Target Ratio |
|---|---:|
| K=1 | 10% |
| K=2 | 20% |
| K=3 | 40% |
| K=4 | 20% |
| K=5 | 10% |

## Actual Distribution

| K-Level | Actual Ratio |
|---|---:|
| K=1 | 5.1% |
| K=2 | 40.7% |
| K=3 | 44.0% |
| K=4 | 9.9% |
| K=5 | 0.4% |

### Observation

The dataset underrepresents:

- Expert-level reasoning
- Multi-step analytical samples
- Deep synthesis tasks

and overrepresents:

- Basic instructional content
- Medium-complexity educational text

---

# 🧪 Synthetic Depth vs Logical Density

The SynD vs LogD scatter analysis reveals:

- Strong diversity in reasoning depth
- Balanced synthetic generation patterns
- Limited clustering artifacts
- Healthy variance across document styles

This indicates robust heterogeneity suitable for pretraining and fine-tuning pipelines.

---

# ✅ Strengths

- Near-perfect deduplication quality
- High structural integrity
- Excellent text cleanliness
- Strong paragraph coherence
- Large percentage of high-quality records
- Robust medium-complexity reasoning coverage

---

# ⚠️ Areas for Improvement

## Increase Advanced Reasoning Data

The dataset contains limited:

- Expert reasoning chains
- Long-form analytical writing
- Scientific derivations
- Multi-hop logical tasks

Recommended actions:

- Add synthetic reasoning traces
- Include theorem proving samples
- Add research-style documents
- Increase chain-of-thought diversity

---

## Improve Complexity Diversity

Target improvements:

| Current | Desired |
|---|---|
| K4 = 9.9% | ≥ 18% |
| K5 = 0.4% | ≥ 8% |

---

# 🚀 Recommended Use Cases

| Use Case | Suitability |
|---|---|
| General LLM Pretraining | ✅ Excellent |
| Educational AI | ✅ Excellent |
| Chat Assistant Fine-Tuning | ✅ Strong |
| Reasoning-Centric Models | ⚠️ Moderate |
| Frontier Reasoning Systems | ⚠️ Needs more K4/K5 data |
| Synthetic Data Generation | ✅ Strong |
| Multilingual Expansion | ✅ Compatible |

---

# 🛠️ Suggested Next Steps

## For Better Frontier-Scale Training

### Add:

- Long chain-of-thought reasoning
- Mathematical proofs
- Agentic workflows
- Research paper synthesis
- Debate and critique samples
- Multi-document reasoning

### Improve:

- Expert-level complexity ratio
- Logical depth variance
- Long-context coherence

---

# 📂 File Structure

```text
project/

├── assets/
│   └── convergeiq_report.png

├── reports/
│   └── convergeiq_results.json

├── README.md

└── LICENSE
```

---

# 📜 Example JSON Output

```json
{
  "ciq_score": 0.8708,
  "dedup_score": 0.9997,
  "struct_score": 0.8602,
  "kdist_score": 0.7561
}
```

---

# 🧩 Metric Definitions

| Metric | Meaning |
|---|---|
| CIQ | Core Information Quality |
| SCR | Sentence Completion Rate |
| PC | Paragraph Coherence |
| DBI | Document Boundary Integrity |
| K-Score | Cognitive Complexity Level |
| JSD | Jensen-Shannon Divergence |

---

# 📖 Citation

```bibtex
@software{convergeiq2026,
  title={ConvergeIQ: Dataset Quality Evaluation Framework},
  year={2026},
  version={1.0}
}
```

---

# 📄 License

This project is released under the MIT License.

---

# ✨ Final Verdict

**ConvergeIQ** reports that the dataset achieves:

- Excellent cleanliness
- Exceptional deduplication
- Strong structural quality
- Good reasoning diversity

The dataset is highly suitable for:

- General-purpose LLM training
- Educational assistants
- Synthetic data augmentation
- Instruction tuning pipelines

However, to support frontier reasoning systems and next-generation agentic models, the dataset should include significantly more:

- Expert-level reasoning
- Long-form analytical synthesis
- Multi-step cognitive tasks
- High-complexity problem solving

---