The Intelligence Layer
The Intelligence Layer is the brain of Flightline. It takes raw execution traces and transforms them into actionable insights and ship-readiness scores. It uses a unique two-tier architecture to balance speed and depth.Dual-Tier Architecture
Flightline does not treat every check the same way. We split our analysis into two tiers to ensure that engineers get immediate feedback while still benefiting from deep semantic understanding.Tier 1: Deterministic (Local)
Tier 1 checks are fast and run entirely on your local machine. They are designed to “fail fast” and catch technical errors like format consistency or pattern violations.Tier 2: Semantic (Cloud/LLM)
Tier 2 checks are reasoning-based and qualitative. They use an “LLM-Judge” to understand the meaning and intent behind the AI’s output, answering complex questions about truthfulness and quality.Evidence-Based Judgments
A core principle of the Intelligence Layer is that every verdict must be supported by evidence. When a check fails, Flightline provides:- The Reasoning: A detailed explanation of why the check failed.
- The Evidence: The specific part of the output or trace that triggered the failure.
- Confidence Scores: A measurement of how certain the Intelligence Layer is about its judgment.
From Verdict to Action
The Intelligence Layer doesn’t just report issues; it suggests how to fix them. Based on the type of failure, Flightline provides specific recommendations to guide the engineer toward a solution.The Overall Verdict
By combining results from both tiers across all 7 Ship-Blocking Questions, Flightline produces an overall verdict of PASS, WARN, or FAIL. This provides a definitive signal for your CI/CD pipeline, allowing you to block broken deployments and ship verified AI with confidence.The 7 Ship-Blocking Questions
Explore the framework that guides the Intelligence Layer’s analysis.