AGENCYSCRIPT
CoursesEnterpriseBlog
đź‘‘FoundersSign inJoin Waitlist
AGENCYSCRIPT

Governed Certification Framework

The operating system for AI-enabled agency building. Certify judgment under constraint. Standards over scale. Governance over shortcuts.

Stay informed

Governance updates, certification insights, and industry standards.

Products

  • Platform
  • Certification
  • Launch Program
  • Vault
  • The Book

Certification

  • Foundation (AS-F)
  • Operator (AS-O)
  • Architect (AS-A)
  • Principal (AS-P)

Resources

  • Blog
  • Verify Credential
  • Enterprise
  • Partners
  • Pricing

Company

  • About
  • Contact
  • Careers
  • Press
© 2026 Agency Script, Inc.·
Privacy PolicyTerms of ServiceCertification AgreementSecurity

Standards over scale. Judgment over volume. Governance over shortcuts.

On This Page

Stage One: EnvelopeThe question it answersHow to apply itStage Two: BoundariesThe question it answersHow to apply itWhen it matters mostStage Three: PriorityThe question it answersHow to apply itWhen it matters mostStage Four: ProofThe question it answersHow to apply itWhen it matters mostPutting the Stages TogetherRun them in orderRe-enter stages as you learnWalking the Framework Through a Real TaskA concrete passWhere each stage saves youAdapting the Framework to Your ContextCompress it for simple tasksExpand it for agent chainsMake it a team artifactFrequently Asked QuestionsWhy start with the Envelope instead of the content?What distinguishes Boundaries from Envelope?Do I always need the Priority stage?How is Proof different from just looking at outputs?Can I skip stages for simple prompts?Does this framework depend on a particular model?Key Takeaways
Home/Blog/A Decision System for Shaping Model Output
General

A Decision System for Shaping Model Output

A

Agency Script Editorial

Editorial Team

·May 9, 2021·8 min read
constraint-based output promptingconstraint-based output prompting frameworkconstraint-based output prompting guideprompt engineering

Most people approach constraint-based prompting by trial and error: write a prompt, see what breaks, patch it, repeat. That works, slowly, but it produces no transferable knowledge. The next prompt starts from scratch. A framework changes that by giving you a repeatable sequence so the lessons from one prompt carry to the next.

Constraint-based output prompting is the practice of defining the exact shape and boundaries of a model's output rather than hoping usable structure emerges. The framework below, which we call Envelope, Boundaries, Priority, and Proof, organizes the decisions you have to make anyway into a deliberate order. Each stage answers a specific question, and skipping a stage tends to produce a specific failure.

Use it as scaffolding, not dogma. The order matters more than rigid adherence to any one step. The value of a named sequence is not that it is magic; it is that it externalizes decisions you would otherwise make implicitly and inconsistently. When the steps have names, you can tell a teammate which stage a prompt is failing at, you can review a prompt by checking each stage, and you can notice when you have skipped one. That shared vocabulary is most of the benefit.

One more reason to prefer a sequence over a checklist of tips: a sequence encodes dependencies. You cannot meaningfully define boundaries before you know the envelope, and you cannot prove anything before both exist. A flat list of best practices hides these dependencies and lets you tackle steps in an order that wastes effort, like writing elaborate test criteria for an output shape you later abandon. The framework's ordering is not arbitrary; each stage produces the input the next stage needs, which is what makes the whole thing efficient rather than merely thorough.

Stage One: Envelope

The question it answers

What is the literal shape of the output? Before anything else, decide the container: a JSON object with named keys, a fixed set of sections, a single value from a closed set.

How to apply it

Write a literal example of the output and put it in the prompt with "match this exactly." The envelope is the foundation; the later stages refine it. Treating output as an envelope with content inside, rather than content with structure sprinkled in, is the reframing that drove the result in What Tightening Output Rules Did for One Support Team.

Stage Two: Boundaries

The question it answers

What must never appear, and what is the allowed range of each field?

How to apply it

State exclusions ("no preamble, no markdown fences") and enumerate closed sets ("priority is one of low, medium, high"). Boundaries are where you prevent the model's default helpfulness from corrupting machine-readable output. The examples in Concrete Scenarios Where Output Constraints Earn Their Keep lean heavily on this stage.

When it matters most

Whenever output feeds another system. For purely human-facing output you can be lighter here.

Stage Three: Priority

The question it answers

When two constraints conflict, which one wins?

How to apply it

Make the trade-off explicit in the prompt: "Prefer accuracy over brevity." Without a stated priority, the model resolves conflicts unpredictably and your output drifts run to run. This stage is the antidote to the conflicting-constraint failure in Seven Ways Output Constraints Quietly Break Your Prompts.

When it matters most

Whenever you have stacked several constraints, especially length plus completeness, which almost always collide.

Stage Four: Proof

The question it answers

How do you know the constraints actually hold?

How to apply it

Build a test set of messy real inputs, define pass criteria, and measure. A constraint you cannot assert against is a hope. The KPIs in Reading the Signal: What to Track When Outputs Must Conform give you the instruments for this stage.

When it matters most

Always, but especially before any production deploy. Proof is the stage teams skip most and regret most.

Putting the Stages Together

Run them in order

Envelope defines the target, Boundaries protect it, Priority resolves its internal tensions, and Proof verifies it. Done out of order, you tune constraints you cannot measure or measure a shape you have not yet defined.

Re-enter stages as you learn

Proof often sends you back to Boundaries or Priority. That loop is the framework working, not failing. Each pass tightens the prompt and teaches you something reusable.

Walking the Framework Through a Real Task

A concrete pass

Suppose you need the model to triage a bug report into a structured ticket. Envelope: define a JSON object with the keys severity, component, and summary, and show a literal example. Boundaries: severity is one of low, medium, high, or critical; component comes from an allowlist; no preamble, output only the object. Priority: if severity and brevity conflict because the report is sprawling, prefer an accurate severity over a short summary. Proof: assemble fifty real reports including vague and duplicate ones, define pass criteria, and measure.

Where each stage saves you

If you skipped Envelope, you would get inconsistent ticket shapes that no script could file. Skip Boundaries and you would see invented components and chatty preambles. Skip Priority and the severity would drift on long reports. Skip Proof and you would ship a prompt that worked on the three clean reports you happened to try and failed on the messy fourth. Naming the stages makes each of these omissions visible before it ships, which is the same discipline the pre-flight list enforces from a different angle.

Adapting the Framework to Your Context

Compress it for simple tasks

For a trivial prompt, the four stages collapse into a few minutes of thought: a one-line envelope, one exclusion, no real conflicts to prioritize, and a handful of test inputs. The framework should not feel heavy on small work. Its weight should scale with the stakes, staying light where failure is cheap and rigorous where failure is expensive, exactly the calibration argued in Choosing How Tight to Make Your Output Rules.

Expand it for agent chains

When constrained output feeds another model call, each stage gets heavier. The envelope becomes an inter-agent contract, the boundaries must anticipate adversarial upstream output, and proof must test the whole chain, not just one link. The framework still applies; it simply demands more at each step because the cost of a leak compounds across the chain.

Make it a team artifact

The framework delivers the most value when a whole team shares it, because then the vocabulary travels. New prompts get reviewed stage by stage, regressions get diagnosed by stage, and onboarding a new prompt author means teaching four named ideas rather than a folklore of tricks. That shared language is what turns scattered prompt craft into a repeatable engineering practice.

Frequently Asked Questions

Why start with the Envelope instead of the content?

Because the container determines everything downstream. If you do not know the exact output shape, you cannot define boundaries, resolve conflicts, or write pass criteria. The envelope is the anchor.

What distinguishes Boundaries from Envelope?

Envelope is the positive shape (these keys, this structure). Boundaries are the negatives and ranges (no preamble, values only from this set). Together they fully specify the output.

Do I always need the Priority stage?

Only when constraints can conflict. A single simple constraint needs no priority rule. But the moment you stack length, completeness, and tone, you almost certainly do.

How is Proof different from just looking at outputs?

Proof uses a fixed, messy test set and written pass criteria, so results are repeatable and comparable. Eyeballing a few clean outputs is exactly the habit that ships fragile prompts.

Can I skip stages for simple prompts?

You can compress them, but do not skip Proof. Even simple prompts benefit from a few adversarial test inputs. The other stages scale down naturally for trivial tasks.

Does this framework depend on a particular model?

No. The stages are model-agnostic. The specific phrasing within each stage may need retuning across models, but the decision sequence stays the same.

Key Takeaways

  • The framework has four ordered stages: Envelope, Boundaries, Priority, Proof.
  • Envelope defines the literal output shape with a shown example.
  • Boundaries state exclusions and enumerate closed sets to protect the shape.
  • Priority makes constraint conflicts explicit so output stops drifting.
  • Proof verifies constraints against messy inputs with written pass criteria.
  • Expect to loop back from Proof to earlier stages; that loop is the method working.

Search Articles

Categories

OperationsSalesDeliveryGovernance

Popular Tags

prompt engineeringai fundamentalsai toolsthe difference between AIMLagency operationsagency growthenterprise sales

Share Article

A

Agency Script Editorial

Editorial Team

The Agency Script editorial team delivers operational insights on AI delivery, certification, and governance for modern agency operators.

Related Articles

General

Prompt Quality Decides Whether AI Earns Its Keep

Prompt quality is the single biggest variable in whether AI delivers real work or expensive noise. The model matters, the platform matters — but the prompt you write determines whether you get a first

A
Agency Script Editorial
June 1, 2026·10 min read
General

Counting the Real Cost of Every Token You Send

Tokens and context windows sit at the intersection of AI capability and operational cost—yet most business cases treat them as technical footnotes. That's a mistake that costs real money. Every time y

A
Agency Script Editorial
June 1, 2026·10 min read
General

Rolling Out AI Hallucinations Across a Team

Most teams discover AI hallucinations the hard way — a confident-sounding wrong answer makes it into a client deliverable, a legal brief, or a published report. The damage isn't just to the output; it

A
Agency Script Editorial
June 1, 2026·11 min read

Ready to certify your AI capability?

Join the professionals building governed, repeatable AI delivery systems.

Explore Certification