CohortDAG

flowchart LR
  A["Import data<br/>CSV, Excel, SAS, SPSS"] --> B["Register tables"]
  B --> C["Build the pipeline on the canvas:<br/>sources, filters, joins, windows,<br/>cohort exclusions, statistics<br/>(live Spark SQL preview)"]
  C --> D["Validate and compile<br/>to deterministic Spark SQL"]
  D --> E["Run"]
  E --> F[("export_data checkpoints<br/>parquet, reused by fingerprint")]
  F --> G["Results: Table 1, Kaplan-Meier,<br/>Cox, incidence and risk ratios, charts"]
  G --> H["Save a portable .csp<br/>with results embedded"]

flowchart TD
  subgraph AG["AI agent — Claude Code, Codex or Antigravity"]
    direction TB
    R["Read the node vocabulary<br/>(list_node_types)"] --> S["Read the live session, read-only<br/>(get_session_state)"]
    S --> W["Author a .csp:<br/>nodes, settings and edges<br/>(review SQL and assumptions)"]
  end
  U["User instruction"] --> R
  W --> V["Validate (validate_pipeline)"]
  V --> P["Compile to deterministic Spark SQL"]
  P --> RUN["Run (run_pipeline / run_statistics)"]
  RUN --> CK[("Checkpoints<br/>parquet, Merkle-reused")]
  CK --> AU["Reference exact audit:<br/>verify_template.py<br/>exceptAll both ways = 0"]
  AU --> Q{"PASS?"}
  Q -- "Yes" --> FR["Freeze verified results:<br/>embed payloads into the .csp"]
  Q -- "No / unresolved" --> ST["STOP, keep artifacts,<br/>diagnose cause, draft a retry"]
  ST --> R
  FR --> HB["Hand back to the human:<br/>launch_app / push_dag (consent-gated)"]