Pre-LLM Prompt Firewall
Problem
Prompts carry hidden instructions, ambiguity, and adversarial framing into model calls.
What breaks
Prompt injection, unsafe completions, non-deterministic behavior.
V1 detects
Injection-style framing, dominance directives, ambiguity carriers, scope drift, constraint absence.
V3 stabilizes
Strip hedges, enforce objective/constraints, normalize instruction structure.
Business model
Sell as a gateway middleware for any LLM product (per-request enforcement).