Load representative data
Status: active
Depends on: start
Incoming edge: start -> load_data
180ms · 1 unit
Typed source table
DAG Execution Simulator
Why explicit workflows beat vague agents in production
This local simulator shows dependency state, failure handling, cost, latency, and downstream readiness. It never executes code or processes private data.
Infer schema
Status: queued
Depends on: load_data
Incoming edge: load_data -> infer_schema
420ms · 2 units
Column map and field types
Plan insight DAG
Status: queued
Depends on: infer_schema
Incoming edge: infer_schema -> plan_insights
910ms · 6 units
Parallel insight tasks
Execute analysis code
Status: queued
Depends on: plan_insights
Incoming edge: plan_insights -> execute_analysis
5.2s · 7 units
Verified tables
Judge verification
Status: queued
Depends on: execute_analysis
Incoming edge: execute_analysis -> judge_verify
2.8s · 10 units
Semantic pass or failure
Generate chart spec
Status: queued
Depends on: judge_verify
Incoming edge: judge_verify -> generate_chart
730ms · 3 units
Chart config
Compile Deck IR
Status: queued
Depends on: generate_chart
Incoming edge: generate_chart -> compile_deck_ir
1.1s · 4 units
Inspectable slide IR
Production lesson
A DAG makes readiness, failure, retries, and risk explicit. If the judge fails, downstream chart and deck nodes stop until the system either retries with a narrower assertion or marks the artifact as risky.
Representative workflow. Customer data, private prompts, internal datasets, and exact costs omitted.