Glossary

ADLC — Adaptive Delivery Lifecycle

The 8-phase delivery cycle that coordinates architecture, design, development, review, testing, deployment, monitoring, and retrospection. Unlike traditional SDLC (linear and human-driven), ADLC is designed for parallel execution with AI agents at each phase.

See: Eight Phases →

AaC — Architecture as Code

The practice of expressing software architecture in machine-readable, version-controlled code (YAML + JSON Schema) rather than diagrams or documentation. Enables AI agents to understand and enforce architectural constraints automatically.

AaC is not documentation. It is a context layer for AI.

MACH-1 — Managed Acceleration of CHange

A framework for controlled high-speed enterprise software delivery combining four layers:

• M — Managed: Quality gates, evidence, audit
• A — Acceleration: ADLC process
• C — Corporate: Architecture as Code
• H — Infrastructure: MCP, CI/CD, fitness functions

Mach Number

Measures how fast you move relative to your change barrier. Not absolute speed — a ratio.

• Mach 0.3: Architecture as code, agents not connected
• Mach 0.5: MCP live, agents read model, manual generation
• Mach 0.7: ADLC partially automated, gates in CI/CD
• Mach 1.0: Full cycle end-to-end, task → production in 1–2 days
• Mach 2.0+: Autonomous agents, predictive analysis

C4 Model

A hierarchical architecture visualization method with four levels: Context (systems), Container (applications/databases), Component (modules), Code (classes/functions). MACH-1 uses C4 for system relationships via LikeC4.

ADR — Architecture Decision Record

A document capturing a significant architectural decision: context, the decision, and consequences. Immutable and versioned. Creates organizational memory of "why we built it this way." Numbered ADR-001, ADR-002, etc.

Quality Gate

An automated checkpoint in the delivery pipeline that verifies a change meets specified criteria before proceeding. Examples: architecture compliance gate, test coverage gate, compliance gate. Each gate produces formal evidence.

Evidence

Formal proof that a change satisfies a quality gate. Not subjective ("looks good") but objective: metrics, rule validation, checksums. Example: "Architecture gate passed: all endpoints match system.yaml api_contracts."

Fitness Function

A quantifiable metric or rule that automatically checks whether code/architecture violates constraints. Examples: "PCI DSS: card numbers must be encrypted," "Layering: service layer cannot call repository layer directly," "Performance: response time < 500ms."

Domain (Business Domain)

A cohesive area of business logic with shared entities, rules, and constraints. Each domain has a YAML spec capturing rules and ontology. Agents read domain specs to generate correct code without manual instruction.

System Spec

A YAML file describing a software system: stack, endpoints, dependencies, business rules, constraints. The single source of truth for AI agents about a system.

Agent

A specialized AI with defined skills, constraints, MCP context, and escalation rules. Six core agents in MACH-1: Architecture, Design, CodeGen, Review, Deploy, Monitor.

See: The Six Core Agents →

Skill

An atomic capability of an agent. Composable (skills combine into workflows) and measurable (each has quality metrics). Examples: "generate_api_contract," "analyze_diff," "check_fitness."

Workflow

A sequence of skills and agents orchestrated to achieve a business outcome. Not a linear checklist — compositions of discrete skills with explicit exit criteria. Example: feature-to-mr workflow chains 6 skill invocations across 3 agents to produce a production-ready MR.

Agent Trigger

The condition that initiates an agent's execution. Examples: a new MR opened (triggers Review Agent), a deployment started (triggers Deploy Agent), a fitness function failure (triggers Architecture Agent).

Escalation Rule

A condition under which an agent stops autonomous execution and hands off to a human reviewer. Examples: confidence score below threshold, regulatory ambiguity, risk tier exceeds policy limit.

MCP — Model Context Protocol

An open protocol for connecting AI models to external data and tools. MACH-1 uses MCP to expose: LikeC4 MCP (C4 architecture), YAML MCP (corporate layer), GitLab MCP (CI/CD and collaboration), custom MCPs (domain-specific tools).

Audit Trail

A complete, immutable log of all changes, decisions, and evidence. Used for compliance reporting and forensics. Captures: who queried MCP, when; what gates passed/failed; who approved exceptions.

Evidence Chain

The complete ordered sequence of evidence records produced during a delivery cycle. An unbroken evidence chain is required for compliance reporting and audit. If any gate fails or is skipped, the chain is broken and delivery is blocked.

LikeC4

An open-source tool for authoring and rendering C4 architecture diagrams from code. MACH-1 uses LikeC4 as the primary format for .c4 files representing system relationships. LikeC4 files are both human-readable and machine-parseable by MCP tools.

White Spot

An area of the architecture not yet covered by Architecture as Code — no system.yaml or .c4 file exists. White spots are tracked and eliminated progressively. Reducing white spots is a key metric for Mach 0.5 → 1.0 progression.