Kevrion Workshop Core

A small TypeScript model derived from verified domain and workflow invariants used in Kevrion, a workshop-management application built around real operational flows.

This repository focuses on organization-scoped relationships, workshop entity invariants, and human-reviewed OCR intake. It intentionally excludes authentication, production database policies, storage infrastructure, and private application code.

Derived from verified domain and workflow invariants used in Kevrion.

What this repository demonstrates

• Every workshop entity carries an organization boundary.
• Access checks compare the actor and entity organization explicitly.
• Customer, vehicle, repair-case, part, and media relationships reject cross-organization links.
• OCR output remains a suggestion until a person confirms the intake data.
• Tests exercise accepted relationships and expected rejection paths.

The implementation is deliberately in-memory and has no network or database dependency.

Domain model

Organization
└── Customer
    └── Vehicle
        └── RepairCase
            ├── MediaReference (photo or document)
            └── Part

Each business entity includes an organizationId. A vehicle links to a customer, a repair case links to its customer and vehicle, and child records link to a repair case.

Workshop workflow

The executable example creates an organization, customer, vehicle, and repair case before attaching a part and a generic media reference. Each creation step validates its parent relationships rather than relying on a user interface to keep them consistent.

The public repair-case statuses group the operational stages into a compact educational model. They are not a copy of the complete application workflow.

Organization boundaries

canAccessEntity and assertCanAccessEntity model organization-scoped access. assertSameOrganization protects relationships, while filterForOrganization demonstrates tenant-scoped reads over in-memory data.

These functions demonstrate domain invariants. They are not Kevrion's production authorization layer and do not claim to prove the security of the full application.

OCR-assisted intake

The OCR model contains only structured, generic fields:

• plate
• VIN
• make and model
• version
• first registration date
• fuel
• power in kW

Each value carries confidence between 0 and 1. The repository contains no image processing, external model call, request template, endpoint, or operational configuration.

All example values are synthetic and were created specifically for this repository.

Human review

OCR output is represented as a suggestion, not an authoritative write. A field can be empty, detected, flagged for review, or manually reviewed. confirmVehicleIntake is an explicit human-confirmation boundary that produces accepted domain data.

Low-confidence fields are surfaced for review; they are not silently discarded. The public model does not pretend that confidence alone proves correctness.

Example

npm install
npm run example

The example completes one valid workshop flow, rejects a cross-organization access attempt, and confirms a synthetic OCR suggestion after a manual correction.

Testing

npm run typecheck
npm test
npm run build

The tests cover same-organization access, cross-organization rejection, relationship consistency, malformed input, confidence validation, manual review, and human confirmation.

Design decisions

Relationships are explicit

Parent-child relationships are checked in domain functions. This makes invalid cross-organization references visible and testable without a database.

OCR is not authoritative

Detected values can be incomplete or uncertain. The model therefore separates machine output, field review, manual edits, and human confirmation.

The public boundary is intentionally small

The repository uses plain TypeScript and the Node.js test runner. It does not need a web framework, database client, AI SDK, or browser runtime.

Security scope

This repository demonstrates selected domain invariants.

It does not contain Kevrion's production authorization, authentication, storage, or database configuration. The in-memory guards illustrate organization boundaries; they are not a substitute for server-side enforcement in a complete application.

What is intentionally excluded

• Authentication, sessions, roles, and invitation flows
• Database schema, row-level authorization policies, triggers, grants, and migrations
• File storage, upload rules, object paths, and signed access
• OCR prompts, model settings, server functions, retries, and provider integration
• Customer, workshop, vehicle, repair, image, and document data
• Deployment, staging, analytics, and incident details

Relationship to Kevrion

The full application uses organization-aware persistence and private media handling. This repository re-expresses a small set of confirmed domain and review invariants as isolated, dependency-light TypeScript so they can be studied without exposing application infrastructure.