# Topic 06 — Language and Terms
## Statistical Analysis of Anomalous Term Distribution (Expanded)

### Purpose of this Topic

**Topic 06** reconstructs and significantly expands upon the prior deep research analysis titled:

> *Statistical Analysis of Anomalous Term Distribution*

The goal is to identify **highly specific language terms** that:
1. Appear in a *focal cluster* of court documents,
2. Also appear in a small number of *related clusters*,
3. **Do not appear anywhere else** in the broader dataset, and
4. Collapse onto a **very small number of documents**, indicating non-organic reuse.

This topic extends the original research by incorporating **structural and metadata connectors** that were not previously available, including tracking fonts and XMP metadata.

---

## Clusters Analyzed

| Cluster Name | Cluster ID | Role |
|-------------|------------|------|
| MATTHEW GUERTIN | 1570 | Focal cluster |
| MUAD ABDULKADIR | 674 | Related cluster |
| ADRIAN WESLEY | 290 | Related cluster |
| PETER LEHMEYER | 696 | Related cluster |

All language terms analyzed here appear **only within these four clusters** and nowhere else in the dataset.

---

## Source of Language Terms

Terms are sourced exclusively from structured language extraction fields:

- `children.doc[].string_raw`
- `children.doc[].string_norm`
- `children.doc[].string_id`
- `children.doc[].string_group_n`

Each row in `children.doc[]` is treated as **one occurrence**.

⚠️ The field `string_quantity` is intentionally **not used**.

---

## What This Analysis Demonstrates

Topic 06 answers the following:
- Which terms are statistically unique to a very small cohort of cases?
- How tightly are those terms concentrated into individual documents?
- Do the same documents also share structural artifacts (fonts, objects)?
- Do language anomalies and structural anomalies converge?

When independent signals converge on the same documents, the probability of coincidence becomes extremely small.

---

## Output Files and How to Read Them

All outputs are written to:

```
reports/06_language_and_terms/
```

### 1. `06_term_cluster_presence_matrix.csv`

Shows where each term appears across the four clusters.

Key columns:
- `in_cluster_*`: presence flags (1 = present)
- `occ_rows_*`: number of occurrence rows
- `doc_count_*`: number of distinct documents

---

### 2. `06_terms_exclusive_to_focal_plus_3.csv`

The core anomalous term set:
- Appears in cluster 1570,
- Appears in at least one of the other three clusters,
- Appears nowhere else in the dataset.

This is the statistical heart of the analysis.

---

### 3. `06_exclusive_terms_doc_concentration.csv`

A document-level leaderboard showing how many anomalous terms appear in each document.

Key column:
- `n_exclusive_terms_in_doc`

This table demonstrates **term collapse** onto a minimal set of documents.

---

### 4. `06_doc_tracking_fonts.csv`

Inventory of documents containing tracking-font-related objects.

Key fields:
- `object_sha256`
- `object_font_name`
- `object_description`

Tracking fonts are structural artifacts independent of language.

---

### 5. `06_exclusive_terms_with_tracking_fonts.csv`

Joins anomalous-term-heavy documents with their tracking fonts.

This table shows **multi-axis convergence** between language and structure.

---

### 6. `06_docs_with_adrian_in_xmp_title.csv`

Documents whose **XMP metadata title** contains “ADRIAN WESLEY” (including known typo variants).

XMP metadata persists across reuse and is a strong linkage signal.

---

## Summary

Topic 06 demonstrates that:
- A small set of terms is exclusively shared among four clusters,
- Those terms collapse onto a very small number of documents,
- The same documents share structural PDF artifacts,
- Metadata further corroborates cross-cluster reuse.

Together, these findings strongly indicate **shared document origin or synthetic reuse**.

---

## Relationship to Other Topics

- Topics 01–03 establish dataset structure and origin.
- Topic 04 establishes clone and artifact reuse.
- Topic 06 bridges **language analysis** with **structural and metadata evidence**.

---

## Intended Use

This topic is designed to be:
- Auditable,
- Reproducible,
- Understandable without SQL,
- Suitable for external review.