#llm TONL structured dataset for LLM reasoning. Follow these rules strictly:
#llm 1. Every non-comment line is part of the data. Do NOT skip sections.
#llm 2. Indentation (spaces) defines nesting and hierarchy of objects/arrays.
#llm 3. Lines like key{a,b,c}: define objects with fields a,b,c.
#llm 4. Lines like key[n]{cols}: define array-tables with n rows and given columns.
#llm 5. Primitive lines use 'name: value'. Values may be bare, quoted, or triple-quoted.
#llm 6. String pool (if present) uses S0, S1, ... references for repeated strings.
#llm 7. Do not hallucinate or invent fields. If data is missing, say it is missing.
#llm 8. Read the ENTIRE file before answering; sections are interconnected.
#llm 9. This is a Markdown document converted to TONL for efficient LLM consumption.
#version 1.0
#delimiter ,
metadata{source_file,parsed_at,format,parser_version}:
source_file: tonl.md
parsed_at: "2026-01-31T10:44:42.607294Z"
format: markdown
parser_version: 1.0.0
structure{total_sections,has_tables,has_lists}:
total_sections: 2
has_tables: false
has_lists: true
sections[2]{level,title,id,content_count}:
2, Methodology, methodology, 0
3, Data Collection, data-collection, 16
section_content[0]{type,value}:
section_content[1]{type,value}:
code, """
Becomes:
"""
paragraph, "section.methodology { section.data_collection { } }"
code, """
---
### 4.2 Paragraphs
- Each paragraph becomes a **raw text block**
- No sentence splitting
- No reflow
- No trimming except trailing whitespace
"""
paragraph, paragraph = "Exact text as written."
code, """
---
### 4.3 Lists
#### Bullet lists
- Become ordered TONL arrays
- Item order is preserved
- Nesting is preserved
"""
code, """"""
list, "[{\"indent\": 0, \"marker\": \"-\", \"text\": \"A\"}, {\"indent\": 0, \"marker\": \"-\", \"text\": \"B\"}, {\"indent\": 2, \"marker\": \"-\", \"text\": \"B1\"}]"
paragraph, "list { item[0] = \"A\" item[1] = \"B\" item[1].list { item[0] = \"B1\" } }"
code, """
---
### 4.4 Code Blocks
- Preserved verbatim
- Language tag (if present) becomes metadata
- Content is never parsed
"""
code, """print(\"hello\")"""
code, """"""
paragraph, "code { language = \"python\" content = \"print(\\\"hello\\\")\" }"
code, """
---
### 4.5 Inline Emphasis
| Markdown | TONL |
| ------------ | ---------------- |
| `**bold**` | preserved inline |
| `*italic*` | preserved inline |
| `` `code` `` | preserved inline |
No semantic elevation is applied.
---
## 5. Explicit Metadata Blocks
Only **explicitly declared metadata** is converted.
Accepted patterns:
- YAML front matter
- Explicit key-value sections
- Declared schema blocks
Implicit metadata is ignored.
Example:
"""
paragraph, "Author: Denis Salvadori Version: 1.2"
code, """"""
paragraph, "meta { author = \"Denis Salvadori\" version = \"1.2\" }"