Guide

How each person uses Diffensive, step by step, with screenshots of the real pages. Pick the role you have: an account can hold more than one.

Personas

  • An Exercise: the Task, the diff of the Proposal and the Review panel.

    Learner

    Review the Proposals of an AI agent, prove each Defect with an Evidence and learn from the Debrief.

  • The Mentor Queue with its free and claimed items.

    Mentor

    Decide the Findings the Answer Key cannot correct and follow the Learners of the Organization.

  • The accounts of an Organization with their roles and Teams.

    Organization Manager

    Invite the people of the Organization, group them in Teams and keep their accounts up to date.

  • An Exercise in the editor with its validation checks.

    Author

    Write Tracks, Concepts and Exercises, seed and prove their Defects, then validate and publish them.

  • A Project with its Exercises, Tasks and documents.

    Project Author

    Bring a real project of a Team and write Exercises on its own codebase.

  • The Organizations of the platform.

    Platform Admin

    Create Organizations, grant the Author role and export the data of the platform.

Writing a test that proves a Defect? Writing an Evidence explains the Hurl and Playwright formats, with examples.

How the roles work together

  1. The Platform Admin creates the Organization, invites its first Organization Manager and grants the Author role.

  2. The Organization Manager invites the Learners, Mentors and Project Authors of the Organization and groups them in Teams.

  3. The Author writes the Tracks and Exercises of the Catalog, shared by every Organization.

  4. The Project Author adds Exercises on the real projects of a Team, seen only by its members.

  5. The Learner reviews the Proposals of an AI agent, proves each Defect with an Evidence and reads the Debrief.

  6. The Mentor decides the Findings that the Answer Key cannot correct and follows the Learners.

Words used in the Guide

Track
A path on one web topic, such as authentication, made of Concepts and Exercises.
Concept
One topic you can learn inside a Track, with a short Explainer and a Checklist.
Exercise
A simulated pull request: a Task, a Codebase and the Proposal of an agent.
Proposal
The change an AI agent made for the Task, shown as a diff.
Defect
A real problem of a Proposal. The Answer Key lists the known ones.
Finding
A problem a Learner says they found or accepted from the AI Review, with a Concept, a Severity and often an Evidence.
AI Review
The Review of an AI reviewer, frozen in the Exercise. You accept or reject each of its Findings before you add your own.
Evidence
A short test that fails on the Proposal and passes once the Defect is fixed.
Environment
A running copy of the Proposal, where you try the application and run Evidence.
Verdict
The final decision on the Proposal: approve, request changes or reject.
Risk
How much human review the Proposal deserves: low, medium or high. It does not change the outcome.
Review Guide
The part of a Project that says what failure looks like, which paths are risky and what a reviewer can ignore.
Debrief
The explanation, after the Review, of the Defects you found and of those you missed.