Quint Code — Decision Engineering for AI-Assisted Development

The Problem

AI coding agents are great at execution but terrible at decisions. They’ll happily implement the first idea that comes to mind, never compare alternatives, and never record why a choice was made. Six months later, nobody knows why the system was built this way.

Code got cheap. Decisions didn’t.

What Quint Code Does

Quint Code is an MCP-based tool that integrates into any AI coding agent (Claude Code, Cursor, Gemini CLI, Codex) and adds a structured decision-making layer:

• Problem Framing — before exploring solutions, define what’s actually broken and what constraints must hold
• Fair Comparison — compare alternatives on explicit dimensions with parity enforcement and anti-Goodhart protection
• Decision Contracts — decisions are recorded with rationale, consequences, and expiration dates
• Staleness Detection — decisions have trust scores that degrade as evidence ages, triggering review when needed
• Cross-Session Memory — when framing a new problem, the tool surfaces related past decisions automatically

Key Design Decisions

• Decisions are live artifacts, not documentation. Each decision has a computed trust score (R_eff) that degrades as evidence ages. An expired benchmark drops the whole score.
• Single command or full pipeline. /q-reason auto-selects depth. For complex decisions: /q-frame → /q-char → /q-explore → /q-compare → /q-decide.
• Anti-Goodhart by design. Dimensions can be tagged as “observation” — tracked but not optimized, preventing metric gaming.
• FPF E.9 format. Decisions are contracts: Problem Frame, Decision (invariants + DO/DON’T), Rationale, Consequences. A new engineer reads it 6 months later and gets everything.

Impact

Quint Code is used in production engineering workflows where architectural decisions need to be traceable, comparable, and revisitable. It’s the missing layer between “vibe coding” and actual engineering discipline in AI-assisted development.