How the SI Score works

SI Score method 2.0

The SI Score estimates how useful a product appears for its stated task from the information available to us. AI reviews published features and limits using a common rubric. Editors can retain their own usefulness rating.

Based on published information. No hands-on task test.

What enters the score

Product usefulness is the entire overall score. We apply an evidence ceiling to the original usefulness rating and publish the lower number. A listing without a recorded usefulness rating and supporting product evidence stays unscored.

Popularity earns no points. Website checks do not affect the total. An unknown price does not reduce it. A high score for one kind of tool does not prove it performs better at another tool's task.

The rating anchors

The writer and checker use the same anchors for every primary task. Each band requires all the conditions in that band. Without support for a higher band, the lower band applies.

• 0–19: no usable task is described.
• 20–39: a task is named, but its workflow is unclear.
• 40–59: a basic workflow is described.
• 60–69: a specific workflow and its audience are clear.
• 70–74: the source explains how the workflow fits that audience.
• 75–79: controls or inputs and outputs are documented too.
• 80–84: the source documents a complete workflow, controls and explicit boundaries or limitations.

The reason for a rating must identify the supported workflow and justify the band. A missing limitation is not proof that the product has none. Scores above 84 are unavailable under this method because we have no controlled task tests.

Evidence limits the score

Current source review: up to 84/100

We have published source text and a usefulness rating made with the current rubric's anchors. The source describes the product; it does not demonstrate task performance.

Earlier or unanchored review: up to 74/100

A recorded rating has source support but lacks the current rubric's anchors. This includes earlier AI ratings and editor ratings without that supporting assessment.

Saved catalogue only: up to 59/100

We could not read the live pages for this review, so the rating uses the saved catalogue record. It does not confirm the product's current features or availability.

These ceilings describe the limits of the evidence. They are not measured confidence percentages. The listing shows the method version, evidence basis and assessment date when recorded. Sources and website checks keep their own dates; a recent page check does not make an older rating new.

An editor's original rating and reason remain saved when AI re-checks a listing. The same evidence ceiling still applies to the published score. Editor approval does not establish hands-on testing.

Website checks are separate

The following diagnostics describe limited checks of the website. They add no points to the product score.

• Basic website checks look for HTTPS, a page response and policy or contact links. A policy link does not verify what a company says or how it behaves.
• HTML accessibility checks look for features such as a page language, viewport, headings and image alternatives. They do not establish accessibility for people using the product, and decorative images may correctly have empty alternatives.
• Homepage HTML response records a single fetch from our server. It excludes full browser loading and does not measure the application's speed.
• Tracker checks look for known trackers in the homepage HTML. They miss later script activity and do not establish how the service handles your data.
• Page date hints come from a copyright year or server date. They do not establish product upkeep or current content.

An unavailable check stays unknown. It does not change the product score.

Facts must pass their own checks

The publication checks compare claims with their sources separately from scoring. A high SI Score cannot make an unsupported listing publishable. Unknown facts stay unknown, and a failed refresh must not overwrite the last accepted listing.

The Human verified label means an editor approved the listing revision. It does not mean the product passed a task benchmark, a security audit or a full accessibility audit.

Use the score with its evidence

Read the stated use case, limitations, source dates and pricing before choosing a tool. Small score differences are estimates from published information. We have not run a controlled, head-to-head task test to establish that one product performs better than another.

Editorial responsibility

Listings are drafted with AI from each site's own pages, then checked against those sources before publishing. Max sets the standards and is responsible for what we publish.

About Max Hall