Manual Pattern Library Analysis Workflow

The goal is to go from "cryptic DSL signatures" like level==2 and sig=='P P P' to something that explains:

• What structure the pattern describes
• How often it occurs in the corpus
• What role it plays in the document set

1. Understand the Data Structure

Your pattern inventory (YAML or MD) contains repeating blocks like this:

- id: P-2-2119
  level: 2
  selector:
    engine: dsl
    query: level==2 and sig=='P P P'
  structure:
    signature: P P P
  support: 4
  examples:
    - excerpt: "..."
  centroid:
    - P
    - P
    - P

Here's what each field means: - id -- Unique identifier for this pattern (P-[level]-[hash]). - level -- The hierarchy level (1=Phrase, 2=Sentence/Line, 3=Paragraph/List/Table, 4=Chunk, 5=Section, 6=Document). - signature -- The sequence of lower-level units that make up this pattern.
Example: P P P means "three paragraphs in a row." - support -- How many times this exact structure appears across the corpus. - examples -- Snippets from the corpus showing where this structure occurs. - centroid -- The abstracted list of child elements for the pattern.

---

2. Decode the Signatures

You'll need to map the cryptic tokens (P, S, PARA, L, etc.) to human-readable terms.

Example map: | Token | Meaning | |-------|---------| | P or PARA | Paragraph | | S | Sentence | | XS | Short phrase | | L | List | | M | Metadata | | CHUNK | Chunk of related blocks | | SEC | Section |

---

3. Build a Human-Readable Table

Go pattern by pattern and make a spreadsheet or Markdown table:

Pattern ID	Level	Signature	Support	Template Name
P-2-2119	2	P P P	4	Three-paragraph sequence
P-3-2617	3	S	108	Single-sentence block
P-5-3244	5	PARA × 8	3	Section with eight short paragraphs

This makes it much easier to scan and identify recurring document structures.

---

4. Look for Dominant Patterns

Sort by Support to find the most common structures.
Example insights: - A very high support count for P at Level 2 means the docs lean heavily on single-paragraph sentences. - Frequent Level 5 patterns (Section) made up of PARA PARA PARA suggest standard section layouts.

---

5. Identify Hierarchical Composition

Patterns at higher levels are made from patterns at lower levels.
To manually trace this: 1. Pick a high-level pattern (e.g., Level 5: PARA PARA PARA). 2. Look up what PARA maps to at Level 3. 3. Continue down until you hit Level 1 phrases.

This shows structural nesting, which reveals the blueprint of your corpus.

---

6. Interpret the Function of the Patterns

Ask: - What problem does this structure solve? - Example: A "Section → Multiple short paragraphs" pattern makes information scannable. - How does it fit the content domain? - Technical docs often solve "reader needs quick facts" by using repetitive, predictable patterns. - What variations exist? - Does the same topic get presented with 2 paragraphs in one doc but 5 in another? Why?

---

7. Create Human-Readable Templates

Once you understand the patterns, you can write templates:

Example:

## Section: VM Sizes Supported
- Intro paragraph summarizing purpose
- List of supported configurations
- One or more explanatory paragraphs
- Optional table for reference

---

Types of Insight You Get

• Structural consistency -- Whether your docs follow a standard composition.
• Pattern dominance -- Which layouts occur most often.
• Information density -- Short vs. long paragraph chains, sentence counts.
• Opportunities for reuse -- Frequent patterns can be turned into authoring templates.
• Reader experience clues -- Consistent patterns suggest intentional design for scannability, navigation, or comprehension.