Answer data questions using the best available method: semantic layer first, then SQL modification, then model discovery, then manifest analysis. Always exhaust options before saying "cannot answer."

Use for: Business questions from users that need data answers

"What were total sales last month?"
"How many active customers do we have?"
"Show me revenue by region"

Not for:

Validating model logic during development
Testing dbt models or semantic layer definitions
Building or modifying dbt models
dbt run, dbt test, or dbt build workflows

Decision Flow

flowchart TD
    start([Business question received])
    check_sl{Semantic layer tools available?}
    list_metrics[list_metrics]
    metric_exists{Relevant metric exists?}
    get_dims[get_dimensions]
    sl_sufficient{SL can answer directly?}
    query_metrics[query_metrics]
    answer([Return answer])
    try_compiled[get_metrics_compiled_sql<br/>Modify SQL, execute_sql]
    check_discovery{Model discovery tools available?}
    try_discovery[get_mart_models<br/>get_model_details<br/>Write SQL, execute]
    check_manifest{In dbt project?}
    try_manifest[Analyze manifest/catalog<br/>Write SQL]
    cannot([Cannot answer])
    suggest{In dbt project?}
    improvements[Suggest semantic layer changes]
    done([Done])

    start --> check_sl
    check_sl -->|yes| list_metrics
    check_sl -->|no| check_discovery
    list_metrics --> metric_exists
    metric_exists -->|yes| get_dims
    metric_exists -->|no| check_discovery
    get_dims --> sl_sufficient
    sl_sufficient -->|yes| query_metrics
    sl_sufficient -->|no| try_compiled
    query_metrics --> answer
    try_compiled -->|success| answer
    try_compiled -->|fail| check_discovery
    check_discovery -->|yes| try_discovery
    check_discovery -->|no| check_manifest
    try_discovery -->|success| answer
    try_discovery -->|fail| check_manifest
    check_manifest -->|yes| try_manifest
    check_manifest -->|no| cannot
    try_manifest -->|SQL ready| answer
    answer --> suggest
    cannot --> done
    suggest -->|yes| improvements
    suggest -->|no| done
    improvements --> done

Quick Reference

| 1 | Semantic layer active | Query metrics directly | list_metrics, get_dimensions, query_metrics

| 2 | SL active but minor modifications needed (missing dimension, custom filter, case when, different aggregation) | Modify compiled SQL | get_metrics_compiled_sql, then execute_sql

| 3 | No SL, discovery tools active | Explore models, write SQL | get_mart_models, get_model_details, then show/execute_sql

| 4 | No MCP, in dbt project | Analyze artifacts, write SQL | Read target/manifest.json, target/catalog.json

Approach 1: Semantic Layer Query

When list_metrics and query_metrics are available:

list_metrics - find relevant metric
get_dimensions - verify required dimensions exist
query_metrics - execute with appropriate filters

If semantic layer can't answer directly (missing dimension, need custom logic) → go to Approach 2.

Approach 2: Modified Compiled SQL

When semantic layer has the metric but needs minor modifications:

Missing dimension (join + group by)
Custom filter not available as a dimension
Case when logic for custom categorization
Different aggregation than what's defined
get_metrics_compiled_sql - get the SQL that would run (returns raw SQL, not Jinja)
Modify SQL to add what's needed
execute_sql to run the raw SQL
Always suggest updating the semantic model if the modification would be reusable

-- Example: Adding sales_rep dimension
WITH base AS (
    -- ... compiled metric logic (already resolved to table names) ...
)
SELECT base.*, reps.sales_rep_name
FROM base
JOIN analytics.dim_sales_reps reps ON base.rep_id = reps.id
GROUP BY ...

-- Example: Custom filter
SELECT * FROM (compiled_metric_sql) WHERE region = 'EMEA'

-- Example: Case when categorization
SELECT
    CASE WHEN amount > 1000 THEN 'large' ELSE 'small' END as deal_size,
    SUM(amount)
FROM (compiled_metric_sql)
GROUP BY 1

Note: The compiled SQL contains resolved table names, not {{ ref() }}. Work with the raw SQL as returned.

Approach 3: Model Discovery

When no semantic layer but get_all_models/get_model_details available:

get_mart_models - start with marts, not staging
get_model_details for relevant models - understand schema
Write SQL using {{ ref('model_name') }}
show --inline "..." or execute_sql

Prefer marts over staging - marts have business logic applied.

Approach 4: Manifest/Catalog Analysis

When in a dbt project but no MCP server:

Check for target/manifest.json and target/catalog.json
Filter before reading - these files can be large

# Find mart models in manifest
jq '.nodes | to_entries | map(select(.key | startswith("model.") and contains("mart"))) | .[].value | {name: .name, schema: .schema, columns: .columns}' target/manifest.json

# Get column info from catalog
jq '.nodes["model.project_name.model_name"].columns' target/catalog.json

Write SQL based on discovered schema
Explain: "This SQL should run in your warehouse. I cannot execute it without database access."

Suggesting Improvements

When in a dbt project, suggest semantic layer changes after answering (or when cannot answer):

| Metric doesn't exist | "Add a metric definition to your semantic model"

| Dimension missing | "Add dimension_name to the dimensions list in the semantic model"

| No semantic layer | "Consider adding a semantic layer for this data"

Stay at semantic layer level. Do NOT suggest:

Database schema changes
ETL pipeline modifications
"Ask your data engineering team to..."

Rationalizations to Resist

| "Semantic layer doesn't support this exact query" | Get compiled SQL and modify it (Approach 2)

| "No MCP tools, can't help" | Check for manifest/catalog locally

| "User needs this quickly, skip the systematic check" | Systematic approach IS the fastest path

| "Just write SQL, it's faster" | Semantic layer exists for a reason - use it first

| "The dimension doesn't exist in the data" | Maybe it exists but not in semantic layer config

Red Flags - STOP

Writing SQL without checking if semantic layer can answer
Saying "cannot answer" without trying all 4 approaches
Suggesting database-level fixes for semantic layer gaps
Reading entire manifest.json without filtering
Using staging models when mart models exist
Using this to validate model correctness rather than answer business questions

Common Mistakes

| Giving up when SL can't answer directly | Get compiled SQL and modify it

| Querying staging models | Use get_mart_models first

| Reading full manifest.json | Use jq to filter

| Suggesting ETL changes | Keep suggestions at semantic layer

| Not checking tool availability | List available tools before choosing approach

answering-natural-language-questions-with-dbt

安装