Semantic Model Lifecycle

Model generation below draws on the catalog, relationships, and metrics that Solid's offline process builds ahead of time — see Collect and Offline Process for how that foundation is produced.

Step-by-Step Lifecycle

AUTO: Model Generation
  └─ Solid auto-generates: tables, columns, joins (incl. many-to-many),
     metrics, fact/dim labels, PKs/FKs, descriptions for every entity
     ↓
HUMAN: Edit Models
  └─ Modeler reviews and edits entities: corrects labels, adds context,
     removes irrelevant columns — before the model can advance
     ↓
AUTO: Benchmark Question Generation
  └─ Solid auto-generates test questions from SQL history; attaches
     certified ground-truth SQL to each question
     ↓
HUMAN: Benchmark Review
  └─ Modeler reviews all questions, adds new ones, removes irrelevant ones,
     updates ground-truth SQLs — before running the test
     ↓
AUTO: Run Benchmark + Score + Root Cause
  └─ Solid runs the benchmark, computes an accuracy score, identifies
     root cause of each failed question
     ↓
HUMAN: Apply Fix Recommendations
  └─ Modeler reviews each recommendation; applies or rejects.
     Solid proposes an exact fix per failure (edit model or add instructions).
     ↓
HUMAN: Approve / Certify
  └─ In Progress → Pending Review → Certified → Exposed to AI Agents
     Only certified models are served to agents via MCP.
     ↓
AUTO: Automatic Updates
  └─ Solid detects schema drift, new columns, changed joins; surfaces
     targeted update recommendations. One-click apply.
     ↓
AUTO: Optimize from Real Usage
  └─ MCP production usage patterns surface coverage gaps; Solid recommends
     additions; modeler reviews and applies.

Approval Lifecycle States

State	Meaning
In Progress	Model created; modeler is editing
Pending Review	Submitted for approval; awaiting certification
Certified	Approved and active; served to AI agents via MCP
Exposed to AI Agents	Only certified models are served; uncertified models are invisible to MCP

The Human-in-the-Loop Principle

Every automated action Solid takes requires explicit modeler review before it takes effect. Solid does not silently update a model. This applies at three points in the lifecycle:

Model generation — the modeler reviews and edits generated entities before advancing
Benchmark questions — the modeler reviews, adds, and edits before running tests
Automatic recommendations — the modeler explicitly approves or rejects each recommendation before any change is applied