Appendix J: The Review Optimization Workflow

Senior Engineers are drowning in “PR Noise.” As AI-generated code volume increases, the “Human Bottleneck” moves to the review phase. This appendix outlines a workflow to preserve human cognitive energy for where it matters most.

The 3-Tier Review System

Tier 0: The Synthetic Sentinel (Automated)

No human looks at the code until the Sentinel Agent passes it.

• Linting & Hygiene: Enforces local standards.
• Security Red-Teaming: Scans for known prompt-injection or credentials.
• The Vibe Check: Compares the new code against the “Local Charter” (Chapter 4).

Tier 1: The “Verification First” Review

The Senior Reviewer does not start by reading the code. They start by reading the Tests.

• The Logic Proof: If the AI didn’t generate 5-10 negative test cases (failure modes), the PR is auto-rejected.
• Traceability: The human verifies that the code matches the “Intent Prompt” committed in the Prompt Journal.

Tier 2: The Taste & Architecture Review (Human)

The human Senior focuses ONLY on Intent and Taste.

• Business Alignment: Does this actually solve the problem we had, or just the problem the AI thought we had?
• Architectural Purity: Does this violate any long-term boundaries?
• The “Alien” Check: Does the code look “Clever but Fragile”?

Senior Reviewer KPI: “Verification Yield”

Stop measuring how many PRs a Senior reviews. Measure their Yield: The number of subtle logic errors or architectural violations they caught that the Synthetic Reviewers missed.

This turns the Senior from a “Janitor” into a “Chief building Inspector.”

“If you review code line-by-line, the bot has already won. If you review the ‘Proof of Logic,’ you are still the master.” — Venkatesh