AI Career OS Refactor: LLM vs Deterministic Split, New ATS-Level Scoring Engine

A major architectural refactor of the AI Career OS project has been implemented, drawing a hard line between generative AI and deterministic logic. The core change is a strict separation of responsibilities: Large Language Models (LLMs) are now exclusively used for content generation tasks like resume enhancement and cover letter writing, while a newly built deterministic engine handles all scoring and consistency logic. This separation is designed to eliminate scoring variability and enforce reliability across the platform's user interface.

The refactor introduces a new, dedicated ATS-level scoring engine. This deterministic module, `src/lib/ats-scoring.ts`, calculates a 100-point 'ATS Fit Score' based on six specific dimensions, including 'Hard Requirements' for certifications and experience thresholds. The explicit goal is to prevent LLMs from ever being used for scoring, ensuring that match scores and their breakdowns are consistent, transparent, and repeatable for every user interaction.

This architectural shift signals a move towards greater trust and auditability in AI-powered career tools. By isolating generative content creation from objective scoring, the project aims to address potential concerns about bias, inconsistency, and 'black box' outcomes in job-matching algorithms. The changes focus on providing users with a clear, reliable metric for their application fit, separate from the AI-assisted content improvements.