What: An optional dedicated hardware system running a local 70B+ parameter LLM that acts as a full QA reviewer — catching the subtle bugs that deterministic tests miss.

What It Does:

Diff Risk Analysis — Reads git diff, reasons about downstream impact, flags high-risk changes with specific line numbers.
Semantic Code Review — Reads the full codebase and checks for logical consistency across C#, JS, HTML, CSS, and seed data.
Headless Browser Testing — Loads pages in Playwright, captures JS console errors, verifies UI renders correctly, screenshots for visual regression.
Test Generation — Generates new deterministic test cases that get committed to the repo, growing the test suite organically.

Hardware Options (New Equipment, 2026 Pricing):

Option	Price	QA Run	Best For
NVIDIA DGX Spark 128GB	~$5K	~10 min	Prove the concept cheaply
Mac Studio M3 Ultra 256GB	~$10K	~12 min	Same ecosystem, silent, simple
RTX PRO 6000 Blackwell 96GB	~$17K	~6 min	Fastest, upgradeable, CUDA
Dual RTX PRO 6000 192GB	~$35K	~4 min	Future-proof, any model

Independent from the deterministic test suite — a factory can run Option A (tests, $0) without Option B (AI), or add it later as a capital investment.

Visual Flow:

flowchart TD
    A[Git diff of changes] --> B[Local 70B+ LLM]
    B --> C[Diff risk analysis]
    B --> D[Semantic code review]
    B --> E[Headless browser: Playwright]
    B --> F[Test generation]
    C --> G[QA Report: pass/flag/fail]
    D --> G
    E --> G
    F --> H[New tests committed to repo]

Feature Roadmap

Node 4 (Primary): Receives Litestream WAL from live VM1. Runs all 8+ analysis modules including procurement scanning. Hosts findings API and local dashboard. Masters analysis.db.
Node 5 (Standby): Receives same WAL stream (second subscriber). Receives analysis.db replica from Node 4 via rsync. LLM model weights pre-loaded. Warm standby — not actively scanning but ready to promote in minutes.