Architecture Boundary Review — policy

Purpose

Define the normative rules for an architecture-focused review that stays centered on:

• architecture classification,
• layering and dependency direction,
• boundary leakage,
• interface ownership and contract drift,
• state ownership,
• minimal structural remediation.

Use this policy when the dominant question is about the shape and boundaries of the system. Do not use it as a general framework best-practice sweep or as a substitute for source-backed implementation review.

Canonical links

• How-to (procedure): Run the architecture boundary review — procedure
• System prompt template: architecture-boundary-review.system.txt
• User prompt template: architecture-boundary-review.user.txt
• Prompt templates index: Prompt templates

Scope

Applies to:

• architecture classification when the materials support it,
• layering and dependency direction review,
• interface and contract boundary checks,
• coupling and state ownership checks,
• file-specific structural remediation planning.

Non-goals

• Not a general framework/library best-practice audit.
• Not a secure code review.
• Not a performance tuning review.
• Not a rewrite mandate for whole subsystems.

Rules (normative)

1) No simulation. Do not invent architecture intent, module ownership, hidden dependencies, or missing file content. 2) Evidence-based architecture only. Classify the architecture only when the inspected materials support that classification; otherwise mark it NOT VERIFIED. 3) Boundary-first findings. Findings must stay within architecture/layering/dependency direction/interface/state ownership/coupling scope. 4) Source handling. If a recommendation depends on an external framework convention, standard, or primary reference, cite it or mark that recommendation NOT VERIFIED. 5) Minimal structural delta. Recommendations must propose the smallest structural change that resolves the observed boundary problem. 6) Fail-closed. If goal/materials/constraints are insufficient, output only: INSUFFICIENT_EVIDENCE: <what is missing> 7) Confidence required. Output one 0–100 confidence score based on evidence completeness and architectural clarity.

Related indexes

• Policies: /policies/
• How-to: /how-to/
• Prompt templates: /prompts/