Ongoing

CADE Controller Package Generator

AI Planning Workflow Designer · 2026 · Core production pipeline · 3 min read

Built an AI workflow that takes a unit OPORD as input and generates the five-document Controller Package that runs a CADE exercise. Every turn, every phase, and every decision prompt traces back to the source order — no drift.

Overview

The OPORD is the source of truth for a CADE exercise. Everything the controller uses during execution — the turn sequence, the situation updates, the decision prompts, the adjudication triggers — has to trace back to it. This workflow takes a unit-provided OPORD, or one I design for the exercise, and uses AI to generate a five-document Controller Package: a master turn list organized by phase, and one turn document per phase with four to five turns each. Controllers and trainers review all documents before execution. If changes are needed, I revise using the model.

Problem

Manually drafting a Controller Package from an OPORD is time-intensive and prone to consistency errors across phases. Turns written independently can contradict each other, drift from the source scenario, or miss operational logic that should carry consequences forward. The governing challenge isn't drafting speed — it's keeping a multi-document output coherent and traceable to one source.

Constraints

  • Every turn must trace back to the OPORD — no independent invention of scenario logic.
  • Consequence state must carry forward across phases, not reset between turns.
  • Controller documents must be usable under time pressure without designer support.
  • All outputs require trainer review before execution.

Approach

I anchored the workflow to the OPORD as the sole source layer. AI generates draft turn content — situation updates, decision prompts, option frames, adjudication triggers — inside that constraint. I review each document for operational coherence and source fidelity before it goes to the trainer team. Trainer feedback comes back to me; I revise using the model and return updated documents. The workflow has produced the Controller Package for all three CADE executions.

AI Operator Skill Demonstrated

  • Used the OPORD as a hard constraint on AI generation — output that contradicted the source was rejected and regenerated.
  • Structured the generation workflow to produce phase documents in sequence so consequence state carried forward correctly.
  • Maintained human review as the gate between AI output and execution-ready documents.
  • Revised based on trainer feedback using the model, keeping the iteration cycle fast without bypassing review.

Key Decisions

Generate the Controller Package from the OPORD, not from freestanding scenario notes

Reasoning:

Freestanding scenario notes drift. An OPORD-quality source layer forces internal consistency and gives reviewers a fixed reference for checking each turn against source intent. Downstream products that contradict the OPORD can be identified and corrected before execution.

Alternatives considered:
  • Draft turn content from a loose scenario summary
  • Let each phase document evolve independently

Keep trainers in the review loop before any document reaches execution

Reasoning:

AI generation produces operationally plausible content, but only trainers can confirm that the exercise logic is sound for this unit in this context. Review is not a formality — trainer changes fed directly into subsequent runs.

Alternatives considered:
  • Deliver documents directly to controllers and collect feedback after execution
  • Run a single design review at the end of production rather than per-document

Tech Stack

  • Claude
  • ChatGPT
  • Markdown
  • OPORD-grounded planning workflows

Result & Impact

  • 5 (master turn list + up to 4 phase documents)
    Documents per package
  • 4–5
    Turns per phase
  • 3
    Executions supported
  • Trainer-reviewed before every execution
    Review gate

The Controller Package Generator made CADE reproducible. The same workflow that produced the first execution produced the third — with trainer feedback incorporated at each cycle. No execution required starting from scratch.

Learnings

  • The OPORD is not just reference material — it is the constraint that makes AI-generated exercise content coherent across a multi-document package.
  • Trainer review is a production input, not a sign-off step. Changes from the first execution improved the workflow for the second.
  • A five-document package generated in sequence preserves consequence logic better than generating all documents simultaneously.
All projects