DecisagentAbout

Decision diagnosis · v0.4

Should AI participate in this decision—and if so, how?

In about 8–12 minutes, turn one decision into a clear recommendation for AI’s role: what it can help with, where a human must stay in control, which evidence is missing, and what to test. Export the assessment for free, then continue into a reviewable Decision Design.

Methodology reviewed · 2026-09-26

Decisagent Fit is not a score or vendor quiz. It separately assesses whether AI participation can be useful and whether consequential action authority is supportable. The result is deterministic and reproducible from user-supplied context that Decisagent has not externally verified; its DA output is a design recommendation, not a product classification or operational assurance result.

  • Free assessment and exports
  • No account
  • No server-side submission record
  • No model call
  • Local Markdown and JSON export
Go to the assessment ↓

About 8–12 minutes

Get an AI-role recommendation you can explain.

Answer six sections about one real decision: its context, evidence, choices, consequences, authority, and outcomes. You will get a recommendation with reasons, evidence gaps, retained human responsibilities, and a next-step Design. Use role names and generalized context rather than confidential records.

Preview the synthetic Fit and Design output first ↓

Synthetic example · actual deterministic output

See what you leave with.

Weekly inventory replenishment
Illustration only · no real orders or outcomes

01 · Fit recommendation

DA-4: Decision-Action Agent

Strong fit. Allow only a specifically approved action through a bounded execution path.

Fallback: DA-3. This is a provisional design recommendation, not proof of a deployed capability or permission to act.

02 · Decision Design

A reviewable plan for the decision

  • 4 explicit alternatives, including an exception path
  • Evidence requirements and freshness rules
  • Human approved action with required human review
  • 5 visible open questions; approvals remain unapproved

Export your own Design Brief, canonical JSON, technical Markdown, or print view for review.

Explore the synthetic Fit → Design walkthrough

This is a read-only synthetic walkthrough. It uses the existing rich replenishment fixture and the same deterministic Fit and Design engines as the free workflow. It does not fill your assessment, create a case, claim a product match, or record a personal Fit completion. The rich example supplies more design detail than a first-time interview normally will.

  1. Frame one recurring decision

    Choose whether and how much synthetic shelf-stable inventory to reorder for each test location during the weekly replenishment cycle.

    Accountable role: Synthetic inventory planning lead. Evidence: Synthetic on-hand, in-transit, forecast, supplier, shelf-life, and policy data are identifiable by item and location.

  2. Keep recommendation and authority separate

    AI can participate usefully if the recommended evidence, governance, authority, and evaluation boundary is implemented.

    The user requested human approved action; external action is assessed separately from whether AI can support the decision.

    The supplied context makes DA-4 the provisional design ceiling. It is a design constraint, not proof that the organization or a deployed product has DA-4 capability.

  3. Turn the diagnosis into a Design

    The synthetic designer adds the real alternatives and policy detail. The deterministic artifact preserves their provenance and leaves required approvals outstanding.

    • Hold current inventory position: Create no purchase order for the item-location pair in this cycle and reassess at the next trigger.
    • Place standard replenishment order: Order a quantity from the qualified primary supplier for standard delivery inside confirmed limits.
    • Transfer existing inventory: Move available synthetic stock from another location rather than purchasing new inventory.
    • Escalate an exception: Route an unresolved shortage, policy conflict, or high-uncertainty case to the inventory planning lead without creating an order.

    Permitted mode, pending accountable approval

    • Prepare a bounded action handoff only after an accountable human records an action-time authorization.

    A visible unresolved question

    How will the implementing team resolve or govern this evidence gap: Cross-system identity, freshness, and reconciliation rules are not yet demonstrated.

  4. Export, then gather evidence

    Your free exports are planning artifacts. A local-first Assurance Case can bind the completed Design to observations, separate accountable gate acceptance, and an explicit authority release. The first-party reference cycle shows how Decisagent used that process for its own v0.9 release; it is a different, observed record, not this synthetic scenario.

    The optional $129 Implementation Pack adds delivery planning. Purchasing it does not accept a Design, pass gates, or grant operational authority.

Read the replenishment decision anatomy → · Inspect the methodology →

Return to your own assessment

What the assessment considers

Fit evaluates eight dimensions separately: decision structure, evidence, objectives and constraints, consequence, actionability, human accountability, outcome observability, and the learning loop. A lower design maturity can be preferable when errors are hard to reverse, accountability must remain human, or outcomes are slow or difficult to attribute. Inspect the methodology in a new tab ↗

How product matches are governed

Decision diagnosis can operate from the accepted Fit methodology, Decisagent’s working decision ontology, and your supplied context. Product matching is stricter: only human-accepted, publication-safe product intelligence can appear. Candidate research is never substituted when the reviewed corpus is empty.