Public method / v1

How we decide.

This is the whole procedure behind a verdict: what gets read, how the answer is chosen, how the number is computed, and what we refuse to publish.

1,177 of 1,222 products have been through it · 45 are earlier imported records still waiting on a run

Stage 1 · the assessment

What goes in, and what comes out.

We fetch three pages for a product - its homepage, its pricing page, and one features or docs page - and search for open-source projects doing the same job. Then the same assessment runs three times over that evidence.

The verdict most runs agree on is the one that publishes, and the product page says how many agreed. Where no verdict wins a majority, the product is not published at all.

Retrieved

Product pageWhat it claims to do

Pricing pageThe listed price, read verbatim

Features or docsWhat it actually does

RepositoriesOpen-source projects already doing the core

Assessed

×3 runsSame sources, repeatedEvery run answers one question: is the useful core realistic to own?

Published

Self-host itIt already publishes its own source

Build itOne session to a usable version

Build the part you useA weekend, with real gaps left

Keep payingThe value is not the code

Stage 2 · the number

The verdict comes first. The score shows its receipts.

The verdict sets a base. From there the number only moves for facts a reader can check on the page. No judgement call moves it, which is why recomputing an assessment always gives the same number.

Verdict base

88–99Self-host it

70–95Build it

40–69Build the part you use

8–39Keep paying

Receipts

+5An open-source build existsSomeone has already published code that does the core job, linked on the page.
+1 / +3Cited sourcesTwo sources earn one point, three or more earn three. Anything the run did not actually retrieve is dropped.
+3Price verifiedThe price appears verbatim on the pricing page we fetched.
+4Every run agreedAll three runs reached the same verdict. A two-of-three majority earns nothing.
−3 eachHard moatsProprietary data, network effects, compliance, infrastructure. Capped at −9.

Evidence score

8–99Printed on every product page with its arithmetic, term by term.

Evidence only moves a verdict down.

Two hard moats, or a build measured in weeks with no open-source implementation to point at, drops an assessment a band. Nothing moves it up.

Self-host overrides the judgement.

Where a product publishes its own source under a licence we confirmed on the repository, the verdict is Self-host it whatever the runs concluded. There is nothing left to build.

The old score was retired for drifting.

Its adjustments were re-chosen by a model on every run, moving the number by up to 68 points over identical sources. Every term above is a recorded fact instead.

Stage 3 · the money

Cost is a separate question.

Everything above asks whether a replacement is realistic. It never looks at your budget. Every product page also carries a calculator that asks the other question - and the two can disagree.

The verdict · stages 1 and 2

Could you own it?

Answered from the pages we retrieved: how much of the useful core is realistic to build and keep running. It never sees what you pay or what your time is worth.

The same answer for every reader.

The calculator · on every product page

Is it cheaper for you?

Answered from numbers you type in: paid seats, what your time is worth, and the hours a build and its upkeep take.

Keep payingprice × seats × 12
Build ithours × your rate + tools + hosting

A different answer for every reader.

Neither answer is an input to the other. A build that costs less can still read Keep paying - cheap is not the same as realistic to own.