Belts · evidence · Randori

Show the work. Test your judgment.

Bring a real artifact, explain your choices and practice adapting when the situation changes.

A review you can prepare today

Use the packet with a person who independently agrees to review your work. It asks for the outcome, the stack you actually used, an artifact, observed checks, a changed constraint and honest limits. It works for a software prototype, a design, a research summary or another field.

Download the blank review packet · Open the packet in your browser

On a phone, open or download the file and edit a copy in your notes or document app. You choose where to keep and share it. The Dojo does not save, receive or send this packet. Browser downloads and printing also work without JavaScript.

Practice, not an award: This packet does not assign a reviewer, score an assessment, issue a belt or certify a professional deliverable. Specialist belts and a formal independent assessment process are still being developed.

How to use the packet

  1. Agree the scope. Choose a field and the proposed practice stage; define the outcome, audience and checks. Agree with an independent reviewer what they may access, share and retain. Disclose conflicts and choose someone with relevant expertise.
  2. Bring the evidence. Record the artifact and source, your role and each agent's role, the harness, model, instructions, versions and permitted tools where known. Include reproduction steps, actual checks, failures and limitations.
  3. Open the work together. The human reviewer checks the original artifact or observes the behavior. For each dimension below, write Observed, Not yet observed or Unclear, with an evidence reference or question. Absence of evidence is not a pass.
  4. Change the situation. The reviewer chooses an unfamiliar constraint suited to the field. The learner chooses an approach, explains it, repeats relevant checks and records what still needs review. Do not share private material with an AI tool without permission.
  5. Leave with a next step. Give one evidence-based strength and one gap; let the learner choose what to do next. Share and retain the review notes only as agreed.

What the reviewer looks for

These are proposed cross-field practice prompts, not a replacement for an existing rank rubric or a passing threshold. Use standards and checks appropriate to the actual field.

Outcome and constraints

The intended audience, result and success checks are clear; unsupported assumptions are named.

Stack and permissions

The learner identifies the harness, model, instructions, tools and permitted actions actually used, with versions when available.

Ownership and delegation

The learner explains their own decisions, what each agent did and why the approach fits the task.

Verification and failures

Inputs, checks and actual results can be inspected; failed or untested cases are recorded rather than hidden.

Adaptation

The learner responds to a new constraint without copying a memorized recipe, explains the change and checks the revised result.

Limits and review

Privacy, permissions and unresolved risks are identified, along with any independent or qualified review needed.

See a fictional review example

A practice sales table has two unique complete lines: 2 items at $4 and 3 at $5. A duplicate is counted once and a missing quantity is left for review. The learner reports $23 and shows the arithmetic: 2 × $4 + 3 × $5. The reviewer independently checks those source rows.

A changed constraint asks for a short text-message report that still names the duplicate and missing quantity. The reviewer records whether the learner adapted without inventing the missing value. This is an illustration, not an observed learner review.

Keep the boundaries clear

The five colored stages on the belts page are proposed field-practice stages, not issued credentials. Developer, Master Developer and Sensei are broader Dojo ranks; a field practice stage does not automatically confer a rank.

For an actual Developer or Master Developer certificate, use the existing Product Package rubric or Master Developer assessment as applicable. A Sensei, or a mentor for Master Developer, reviews; a separate Maintainer decides under the certificate process. Sensei selection follows the existing process. This practice packet does not change those rules or promise a review.

For regulated or safety-critical work, a practice review is separate from independent validation and any approval required from a qualified professional.

Review notes, scores and learner work are private by default. Ask before sharing learner material with an AI service; a human reviewer reads the evidence and writes feedback. See the Sensei privacy guide and review guide.

Choose a Randori challenge · Return to belt practice