Auto-Maintain
How Solid detects schema drift, benchmark failures, and production usage gaps, and proposes one-click fixes to keep semantic models current.
Auto-Maintain is the third of Solid's three loops — keeping semantic models current as data and usage evolve.
Trigger Type A: Benchmark Run Failures
When: A benchmark run returns failed questions.
What Solid does:
- Identifies the root cause per failure (ambiguous question, wrong join, missing metric, etc.)
- Proposes the exact fix for each failure: edit a specific part of the model, or add a specific instruction
- Presents each proposal as a one-click "Apply Fix"
Human gate: The modeler reviews each proposal and applies or rejects it individually.
Trigger Type B: Data Layer Changes
When: Tables, columns, metrics, or SQLs evolve in the warehouse — DDL changes, new columns, renamed fields, deprecated tables.
What Solid does:
- Automatically detects the change by re-pulling schema/metadata on a configurable schedule (daily/weekly)
- Maps the impact to all affected semantic models
- Surfaces a targeted update recommendation per model — e.g., "Column X was renamed to Y — update these 3 references"
Human gate: The modeler reviews each recommendation and applies or rejects it.
Trigger Type C: Production Usage Gaps
When: Real MCP agent usage reveals a metric or concept with no model coverage — agents are asking about things no semantic model currently addresses.
What Solid does:
- Flags the gap with full context (what was asked, how often, which model was closest)
- Suggests recommendations based on usage patterns
Human gate: The modeler reviews and decides whether to extend an existing model or create a new one.
Schema Drift Detection Mechanics
Solid pulls schema/metadata on a configurable cadence (daily or weekly). It processes DDL deltas — not full re-ingestion — to identify what changed. The diff is presented to modelers with impact mapping: which models are affected, and which specific entities need updating.
Updated 8 days ago