AIML Integration: Predictive Prefetch & Forensic Diagnostics

1. Overview & Architectural Role

In 17-cdn-caching-proxy, AI capabilities are partitioned into two distinct operational tiers:

2. Tier A: Popularity Predictor (Proactive Prefetching)

Specification: aiml/MODEL_CARD.md
Objective: Evaluates historical URL request distributions (Zipf popularity curves) to forecast cache misses before they occur.
Evaluation: Evaluated against a non-AI static baseline in aiml/eval/evaluate_model.py.
Graceful Fallback: If inference latency exceeds 5 ms or confidence is below 80%, the proxy defaults entirely to standard reactive LRU caching.

3. Tier B: Cache-Miss Forensic Diagnostics

Objective: Analyzes origin headers, request inspector ring buffers, and cache-miss patterns to identify misconfigured Cache-Control directives or missing Vary headers.
Prompt Safety Guardrails:
- LLM outputs are validated against strict Zod JSON schemas.
- The model is never on the synchronous request-response path.
- Zero private user tokens or cookies are included in forensic prompt contexts.