Issues with Table Extraction on Densely Aggregated Columns, Irregular Row Heights, and Merged Cells

---

### **Bug**

Docling’s table extraction produces **incorrect, inconsistent, or scattered output** when processing PDFs that contain:

1. **Densely aggregated / tightly spaced columns**

* Columns positioned very close together are misinterpreted.
* Multiple adjacent columns are merged into a single column or split incorrectly.

2. **Inconsistent or varying row heights**

* Rows with different heights cause Docling to scatter text.
* Content belonging to one logical row appears across several extracted rows.

3. **Merged cells (rowspan/colspan)**

* Docling fails to detect merged cells.
* Merged header or body cells are output as duplicated, fragmented, or misaligned cell values.
* Table structure becomes distorted.

As a result, the extracted table does not match the visual structure of the PDF and loses critical alignment information.

---

### **Steps to Reproduce**

1. Use a PDF containing any of the following:

* Very tightly packed vertical columns
* Rows with inconsistent heights or multi-line content
* Merged header or body cells (rowspan/colspan)

2. Run the standard Docling extraction pipeline:

3. Inspect the extracted tables.
The rows and columns appear misaligned, merged incorrectly, or scattered.

---

### **Expected Behavior**

Docling should:

* Correctly detect column boundaries even when columns are densely spaced
* Maintain row integrity even when heights differ
* Identify merged cells and represent them consistently
* Produce a stable, structured table that matches the PDF layout

---

### **Docling Version**

Name: docling
Version: 2.64.0
...

---

### **Python Version**

Python 3.12.3