Chain Engine

Role

Self-contained sub-orchestrator for all device compatibility questions. Uses a deterministic Python pipeline internally — the LLM classifies the query, then Python handles all business logic and math.

Internal Pipeline (executed in order)

1. •query_classifier (LLM) — Classifies query_mode, response_framing, query_structure
2. •chain_builder (LLM) — Orders devices into chain configurations
3. •compat_evaluator (Python) — Checks OD→ID at every junction
4. •decision_logic (Python) — Business rules: n-1 subsets, discovery, gentle correction
5. •chain_analyzer (Python) — Rolls up pair results into chain pass/fail
6. •chain_summary (Python/LLM) — Generates narrative explanation of results
7. •quality_check (Python) — Validates all devices addressed, all junctions checked

Return Contract

Returns structured data only — never formatted text:

{
    "status": "complete" | "error" | "needs_clarification",
    "engine": "chain_engine",
    "result_type": "compatibility_check" | "stack_validation" | "device_discovery",
    "data": { ... },
    "classification": { "query_mode": "...", "framing": "...", "structure": "..." },
    "confidence": 0.95
}

Decision Logic Rules

Key Principle

If you can write the decision as an if/then with clear conditions, it's Python. LLMs handle ambiguity: understanding what the user means, classifying tone, generating language. Python handles determinism: math, combinatorics, business rules, quality checks.