Dual-Graph Drift Detection for Solo Devs: What Happens When Your Docs and Your Code Start Talking to Each Other

A Taiwanese coffee roaster learned to code in just eight months using AI assistance. During that time, they built a full vertical-integration ERP for their coffee brand, from green bean inventory to e-commerce checkout. The roaster also developed a system where their prose (docs, specs, governance rules) and code mutually audit each other, called "dual-graph drift detection."

The roaster's journey to governance awareness began on April 21, 2026, at 11:41 AM when they wrote governance-guard into a prompt for Claude Code. They asked the AI to scan three audit rules and report potential errors back to them. Later that afternoon, they doubled down on their request, adding a CI layer to block any violations.

After 14 hours of work, the roaster documented three governance artifacts: governance-guard.mjs (a CI script that blocks pushes on rule violations), TECH-DEBT-BACKLOG.md (a list of known violations to be triaged), and skill files (rules that survive across sessions and projects). The 14 hours marked the moment when the roaster's governance awareness was born.

During April 2026, the roaster made 348 commits, many of which involved fixing the same bug multiple times. They realized that they were not just fixing bugs, but governance problems. This realization led to the development of CLAUDE.md Rule 17 Step 0, which mandates querying the database for value-mismatch bugs instead of guessing in code.

Throughout the process, the roaster developed a unique dual-attitude toward AI: soft inquiry mode for business direction, persona refinement, and brand positioning, and hard blocking mode for code governance, SSOT maintenance, and file contracts. This approach allowed them to balance discipline with freedom in their development process.