Methodology
How every verdict on this site is calculated
A verdict is never typed in by hand. A human fills in ten sub-scores, each one backed by a link. The two totals and the verdict come out of a formula that lives in version control, so the same inputs always produce the same answer.
1. Vibe code score (0–10)
How realistically an AI coding agent can rebuild the product. Weighted average of four sub-scores:
| Factor | Weight | The question asked |
|---|---|---|
| UI simplicity | 35% | How trivial is the interface to generate? 10 = a form and a table. |
| Logic triviality | 30% | CRUD and business rules score 10. A proprietary ranking algorithm scores 0. |
| Data independence | 20% | Is it useful on day one with no third-party dataset? 10 = yes. |
| Integration surface | 15% | How few integrations does it need before it is useful? 10 = none. |
2. The two hard caps
Some products have a genuinely simple interface and are still impossible to replace. Two flags force the vibe code score down to 3, whatever the sub-scores say:
- asset_backed — the product owns warehouses, fleets, hardware or stock. You cannot vibecode atoms. A 3PL dashboard is easy; the 3PL is not.
- regulated_entity — the product holds a licence or carries legal liability: payment institutions, insurers, lenders, customs brokers. You cannot vibecode a licence.
3. Moat score (0–100)
How hard the company is to displace, independently of how hard the software is to write. Straight sum of six components:
| Component | Max | The question asked |
|---|---|---|
| Proprietary data | 25 | Data nobody else can collect. |
| Network effects | 20 | Does each new customer make it better for the others? |
| Regulation | 20 | Licences, certifications, legal liability. |
| Integration lock-in | 15 | How deeply it is wired into everything else. |
| Switching cost | 10 | Migration pain, retraining, historical data. |
| Contracts | 10 | Multi-year deals, procurement, SLAs. |
4. The verdict rule
if (moat > 70 || vibe <= 4) -> NOT REALLY if (vibe >= 7 && moat <= 40) -> YES if (vibe >= 5 && moat <= 70) -> KINDA otherwise -> NOT REALLY
A deep moat always beats a simple interface. A product can be trivial to build and still get a NOT REALLY, because the question is not "can this be written?" but "can you actually stop paying for it?".
5. Evidence and provenance
- — Every price shows the plan it refers to, the date it was captured, the assumptions behind it and a link to the vendor's own pricing page. Prices older than 90 days are flagged as needing a recheck.
- — Every product page is signed by a named human with a review date and the sources consulted.
- — A linter runs over the whole catalogue and blocks publication on contradictions: a verdict that does not match the formula, a cap that was ignored, a price with no source.
- — Vendors have a right of reply. If a vendor disputes a verdict, their response is published on the page, unedited.
6. Bias and money
Sponsor slots are clearly labelled and buy placement, never a verdict. No sponsor has ever seen a score before publication, and buying a slot does not change one.
The Lovable link in the footer is an affiliate link. No product analysed on this site is monetised through affiliate links.
Declared conflict of interest. The author is the official representative of Magento Open Source for Italy at Netcomm and works at Host S.p.A., which sells Magento hosting. Read the Platforms and Hosting & Infra entries with that in mind.