How ChatPDM Works

ChatPDM processes every query through a fixed deterministic pipeline.

Flow

input → normalize → classify → validate → resolve / refuse

Steps

1. Normalize

• standardizes input text
• removes superficial variation

2. Classify

• determines query shape
• examples:
  • exact concept
  • subtype
  • comparison
  • relation

3. Validate Constraints

• checks if the query fits authored rules
• blocks unsupported composition
• internal enforcement layer (referred to in code as "ZeroGlare")

4. Resolve or Refuse

• resolve if a valid concept mapping exists
• refuse if:
  • no exact match
  • out of scope
  • invalid composition

Output

Every response follows a fixed contract:

• type
• query interpretation
• resolution method
• suggestions (if applicable)

No step introduces new meaning at runtime.