ReputeCheck

The problem

A rating out of five explains nothing

A star rating tells you a number and hides everything that produced it. You cannot see how many reviews it rests on, whether the people leaving them ever bought anything, whether the business is registered, or whether the parcel actually arrived. Two businesses can show the same 4.6 when one has earned it over four hundred verified purchases and the other over nine reviews from accounts created last week.

That gap matters most exactly where trust is thinnest — buying from a page you found an hour ago, paying in advance, hiring someone you have never met. Bangladesh runs a great deal of commerce on precisely those transactions.

ReputeCheck's position is that a score is a claim, and a claim owes you its evidence. Every score on the platform carries the inputs that produced it, a confidence level reflecting how much evidence exists, and reason codes naming what pushed it up or down — and the subject of a score can contest it.

What it does

Everything needed to build trust — and to contest it

Four independent evidence streams feed one score. They are deliberately separable: a business can be genuinely registered and still deliver badly, and the score should be able to say so.

The product

Bilingual by construction, not by translation layer

English and বাংলা are both first-class: the content model stores each language, rather than machine-translating one into the other at render time.

Where it stands

Live infrastructure, early network

ReputeCheck is deployed and serving across six hosts: a Nuxt 3 server-rendered public site, a Laravel API, and four separate single-page panels for consumers, businesses, developers and administrators.

The content on it is still seed data — 189 subjects, of which the overwhelming majority are deliberately unclaimed and carry no score at all. That is the honest state of a trust network before its network exists, and inventing activity would defeat the entire premise of the product.