Orders Production

A three-stage AI pipeline that produces exercise-ready OPORDs for brigade and below — governed by doctrine, adversarially reviewed, and gated before unit release. The goal is not a perfect order. It is a 90% first draft that saves units significant prep time.

Key Design Decisions

Build a pipeline, not a prompt

Reasoning

Prompts produce plausible text. A governed pipeline — with writing guide guardrails, adversarial review, and a formal quality gate — produces a product the unit can actually use. The constraints are what make the output credible.

Alternatives considered
  • Prompt engineering with increasingly detailed system prompts
  • Single-pass generation with manual human review

Separate red-team as a cross-project skill

Reasoning

Skeptic-based adversarial review is not specific to orders. Keeping it separate makes it reusable across any AI-produced artifact — plans, briefs, exercise materials. The orders pipeline calls the red-team skill; the red-team skill does not depend on orders.

Alternatives considered
  • Embed skeptic sweep directly in opord-writer
  • Run adversarial review only within order-release

Target 90% first-draft quality, not 100%

Reasoning

A 100% ready order is not an achievable or honest target for AI generation at the current state of the art. A 90% first draft that has been validated and gated saves the unit the most time — they spend prep time on decisions, not structure.

Alternatives considered
  • Present the output as complete and require unit to identify gaps
  • Require full human rewrite of AI output before delivery
Back to overview