Testing Methodology — How We Evaluate AI App Builders | Markido

How Lovable.markido.one tests AI app builders — the exact tasks, evaluation criteria, and scoring approach used for every comparison on this site.

Why a Published Methodology Matters

Most AI tool review sites publish ratings without explaining how they were generated. A "9.2/10" is meaningless without knowing what was tested. This page documents the exact process used to evaluate every tool covered on this site so readers can judge whether the methodology matches their actual use case.

Standard Evaluation Taskset

Every tool is evaluated against the same four-task sequence. These tasks represent the complete workflow for a non-technical founder who wants to ship a working product using an AI app builder.

1. Task 1 — Full-stack generation from a single prompt:

   Prompt used: "Build a multi-tenant SaaS dashboard with user authentication, a data table showing mock usage statistics per workspace, and a settings page. Use Tailwind CSS for styling."

   Evaluated on: Does the tool generate a complete React frontend? Does it create a database schema? Does it scaffold authentication? Is the result deployable without additional code?

2. Task 2 — Payment integration via follow-up prompt:

   Prompt used: "Add Stripe payment integration with a subscription billing page that shows the current plan, allows upgrade/downgrade, and handles webhook events for subscription changes."

   Evaluated on: Does the tool generate payment intent creation code? Does it create a webhook handler? Does it update the UI correctly? Does it handle error states?

3. Task 3 — Code export and ownership:

   Action: Export the complete project to a GitHub repository using the tool's built-in export feature.

   Evaluated on: Is a full code export available? Is the code readable by a human developer? Are dependencies listed in package.json? Is the export on the free or paid tier?

4. Task 4 — Custom domain deployment:

   Action: Deploy the generated app to a custom domain using the tool's built-in deployment feature.

   Evaluated on: Is custom domain deployment available? What plan is required? How many steps does deployment take? Does the deployed app work correctly?

Scoring Dimensions

Testing Environment

• All tests run on a fresh account with no prior history on the platform
• Each tool is tested on its own terms — we do not penalise a design tool for lacking a database if it does not claim to provide one
• Pricing data is verified directly on the vendor's pricing page at time of testing
• Tests are repeated after major platform updates; the date of the most recent test is noted on each comparison page
• We do not use vendor-provided test accounts, pre-configured templates, or promotional credits

Limitations

No methodology is perfect. The taskset above is optimised for non-technical founders building SaaS or internal tool MVPs. It does not capture:

• Mobile app generation (iOS/Android native) — evaluated separately for tools that support it
• Enterprise-scale performance — we test at prototype scale, not thousands of concurrent users
• Design quality at the pixel level — we assess functional correctness, not visual polish, unless the tool's primary claim is design quality
• Long-term reliability — tests are point-in-time; platform stability over months is noted qualitatively when we have data

Affiliate Relationship

Lovable.markido.one earns affiliate commissions from Lovable.dev. This relationship was established after Lovable.dev scored highest in the evaluation, not before. The methodology above was applied to Lovable.dev identically to all other tools. The affiliate commission does not affect ratings. Full details: affiliate disclosure →

Questions about the methodology? See the author bio →