Diagram Contract & Architectural Anchoring
Purpose: Rigorous architectural illumination contract mandating referentially anchored diagrams, falsifiable claims, and sequence failure paths across major subsystems.
| Specification Metadata | Value |
|---|---|
| Framework Version | CEF v0.1.0 |
| Governance Tier | Authoritative Core (Evidence Standard) |
| Target Roles | Lens Specialists, Adversarial Auditors, System Architects |
| Foundational Standards | Simon Brown's C4 Model, Mermaid Specification, ISO/IEC 25010 Understandability |
1. Mandatory Architectural Invariants
In the Codebase Evaluation Framework, diagrams are reproducible evidence, not cosmetic decoration. Every diagram must make non-obvious runtime structures legible and must be referentially anchored to concrete symbols within the codebase.
Diagrams are required under the following conditions:
| Evaluation Scope | Lifecycle Phase | Mandatory Diagram Types |
|---|---|---|
| Repository Truth Map | Wave 1 | System Context (C4-L1) and Container Architecture (C4-L2). |
| Major Subsystems | Wave 1+ | Component Architecture (C4-L3) or Module Dependency Graph. |
| Cross-Service Protocols | Waves 2–3 | Paired Sequence Diagrams: Happy Path and Fail-Closed / Error Recovery Path. |
| Non-Obvious Algorithms | Waves 1–3 | State Machine Transition Diagram or Algorithmic Flowchart with rationale. |
| Concurrency & Locks | Wave 2 | Interaction Sequence or Collaboration Diagram identifying locks and queues. |
Major Subsystem Definition: Any package, module, or directory owning a distinct architectural responsibility (such as storage, API, scheduler, consensus, authentication, or UI) with >1,000 lines of code or explicit domain boundaries.
2. Referentially Anchored Metadata Contract
To prevent speculative abstraction and model hallucinations, every diagram file must include an authoritative YAML metadata frontmatter block:
diagram_id: D-STORAGE-01
type: c4_container
title: "Content-Addressable Storage and Write-Ahead Log Subsystem"
anchors:
- path: pkg/storage/file_object_storage.go
symbol: WriteObject
note: "Primary write path committing CAS blobs"
- path: pkg/storage/wal/wal.go
symbol: AppendEntry
note: "Write-ahead journal serialization"
claims:
- "Writes are committed to the write-ahead journal before content indexing."
- "Index lookups fail closed on cryptographic hash mismatch."
evidence_grade: E2
Verification Criteria
- Concrete Symbol Anchors — All declared anchors must resolve to physical file paths and exported symbols in the audited commit. Unanchored diagrams remain at evidence grade
E1and cannot proceed to handoff. - Falsifiable Claims — Every diagram must state explicit, falsifiable architectural assertions. Adversarial auditors specifically test and challenge these claims.
3. Supported Notation & Diagram Technologies
Evaluators must adhere to the following technological priority order:
- Native Mermaid in Markdown — Standard format for portable, pull-request-reviewable architectural diagrams rendered natively across documentation portals.
- Declarative D2 Specifications — Preferred for complex dependency graphs and topological layouts.
- Structured ASCII Art — Fallback representation when automated rendering engines are unavailable in execution environments.
- External Modeling Engines — PlantUML or Structurizr may only be used when explicitly authorized by the operator; otherwise recorded as a tooling gap.
4. Sequence Diagram Invariants
When modeling inter-process, network, or asynchronous workflows, sequence diagrams must satisfy the following minimum criteria:
- Literal Entity Naming — Actor, service, and component names must correspond exactly to structs, packages, or processes in the source code.
- Explicit Protocol Messages — Request, response, and event labels must reflect real API endpoints, method signatures, or channel types.
- Mandatory Failure Paths — Every happy-path sequence must be paired with an explicit error, timeout, or cancellation flow.
- Concurrency & Timeouts — Diagrams must indicate explicit timeout values, retry limits, and mutex boundaries documented in the implementation.
5. Architectural Tradeoffs & Anti-Hallucination Controls
| Advantage | Consideration & Risk Mitigation |
|---|---|
| Forces evaluators to inspect real implementations rather than imagining high-level boxes. | Mandatory AST anchor validation prevents synthetic architecture generation. |
| C4 modeling scales consistently from high-level context down to individual components. | Evaluators must focus on complex hotspots to prevent budget exhaustion on trivial CRUD handlers. |
| Explicit fail-path sequences systematically expose reliability and error recovery vulnerabilities. | Diagram metadata includes commit Git SHA to prevent drift across continuous refactors. |