How The Comparator works

Capability: what a product can do

Capability is read from the product itself. It is built from:

• accepted product facts — specifications and identity we have verified or accepted;
• deterministic derived dimensions — values computed the same way every time from those facts (for example an upscaling grade derived from a graphics architecture);
• scenario rules — a fixed rubric per job (1440p gaming, local AI, system drive, workstation board…);
• a methodology version — the rules in force, listed under Versions below.

Price, the store, availability and affiliate status never change a capability grade. A product with no offer today keeps its grade.

Scenario grade: S, A, B, C, D, F

A grade is not the quality of a product in general. It is its suitability for one defined task under one methodology. The same graphics card can be S for 1440p gaming, B for ray tracing at 1440p and C for local AI — that is the point, not a contradiction.

S ExceptionalA ExcellentB GoodC UsableD LimitedF Not suitable

Today, for example: RX 9060 XT 16GB — 1440p Sweet Spot A · Ray Tracing (1440p) C · AI / LLM B.

A dash (—) means the facts that scenario needs are missing: it is not an F, and no substitute value is guessed for the grade.

Rank: a place inside a defined universe

#7 is the product's place among the products ranked for that scenario — grade first, then measured capability — not its place among all hardware. Each scenario ranks its own universe:

• identity: only products with a trustworthy canonical identity hold a public rank; a listing that does not resolve to a known product can be sold, but is never ranked;
• segment: a workstation graphics card is ranked for AI and rendering, not for gaming; server memory is not ranked for a gaming PC;
• platform: motherboards rank within their socket — an AM5 board never competes with an LGA1851 one, because one can never replace the other;
• facts: a product without the facts a scenario needs is shown with a dash, outside that scenario's ranking.

Value: what you get for the money, under your priorities

Value = f(capability, current comparable price, your priorities, constraints, market, methodology version)

Value is not a property of the product. It is computed for a view: the products in it, their prices today and the priorities you chose. It can change when:

• the price changes — capability stays the same;
• the priorities change — the order of the list can change with them;
• the market changes — a rival gets cheaper, a product enters or leaves;
• the view changes — filters and constraints decide which products are compared.

Value 100 means the top end of the current normalised scale for this view — under these priorities, at today's prices, among the products being compared. It does not imply a unique winner: more than one product can reach 100. Two pages can show different Values for the same product because they compare different products or use different priorities.

Presets are starting points

Balanced, Gaming, Productivity, AI and the other presets are starting configurations of the weights, not recommendations. Change any weight and the profile becomes Custom. The priorities you choose in a category are remembered in this browser and used by the other pages of that category where they can (the Tier list's Value column, for example).

Try it: open the graphics-card market with the 1440p preset and move a weight — the Value column and the order change, the capability grades do not.

Your own price and your own deal

A price you enter — a used listing, a deal in a store we do not track — is evaluated against the same current market and the same priorities as everything else. It stays yours: it does not become a tracked market offer, and it does not change anyone else's numbers, unless you explicitly save it as your own data in this browser.

Electricity and total cost of ownership

When you switch total cost of ownership on, an estimate of the electricity a product uses over three years, at the rate you set, is added to the cost side of Value. It can change Value and the order; it never changes capability.

What “current market” means

Everything priced on the site comes from one market at a time — here United States, in USD. A market snapshot is the set of offers accepted into that market: the products they belong to, how many offers there are, and how fresh each source is.

How fresh a price is

source refresh → accepted observation → market snapshot

• When a source is refreshed and an offer is accepted, its price counts as checked at that refresh — even if the number did not change. A price is not “old” because it stayed the same.
• Sources are refreshed at different times. The snapshot has an overall date, and the source freshness line says how many sources were refreshed within the last day; we do not claim every store was checked at the same second.
• Dates are stored as machine-readable UTC timestamps and shown as dates; where a source is stale, the page says so.
• The retailer's page is always the final word on price, stock, shipping, tax and terms.

Current and last observed offers

Current offer → No current offer → Legacy product

A product with no current offer is still a product: its facts, capability grades and ranks stay, and its page explains what happened to the market. It shows as “no current offer” when we observed an offer within the last 90 days, and as a legacy product after that or when we never did. Discontinued is stated only on evidence, never inferred from a missing offer.

Example: Arc A310 — no current tracked offer, with its capability, its last observed price as history, a check for a price you found and upgrade options.

Which offer sets the comparable price

One model can have many offers. Market calculations use one eligible comparable offer per product, chosen by a versioned offer policy from properties of the offer and its source — never by a retailer's name:

• the right market;
• the intended condition (new, for the comparable price);
• a source whose price is comparable and eligible for Value;
• a dated, verified, fresh price;
• a listing that resolves to the right product.

Among eligible offers the lowest landed price (price + shipping) wins. Where to buy can show more offers than the one used for comparison. The policy is versioned like everything else (see Versions).

Product identity

10 offers ≠ 10 products. The Explorer shows Products by default (one row per product, its best eligible offer) and All offers when you want every listing.

Offers are matched to products by product codes and verified identity, not by guessing from a name. Duplicate catalogue records of one product are merged (the boxed and tray versions of a processor are one processor); a listing that cannot be matched stays out of rankings until it is resolved.

Facts and provenance

• Official and manufacturer facts where possible;
• retailer listing facts where they are the appropriate source (condition, price, shipping);
• deterministic derived facts, computed the same way every time;
• estimated values are marked as estimates, with where they came from;
• conflicting sources are withheld rather than guessed.

Unknown stays unknown: a missing fact is shown as — / Unknown, not invented, and “we do not know” never counts as “it does not have it”. Missing data must not become fake precision.

Benchmarks: what they are here and what they are not

A capability grade is not a benchmark score, and The Comparator does not present its indexes as its own laboratory results.

One system, six views

Every page uses the same facts, capability grades, market snapshot and priorities. They differ in the question they answer:

What a Best page means

Best pages use deterministic published criteria, shown on the page: the products eligible, the capability each one must reach, the priorities used for Value, and the market. Within those criteria they list the highest current matches.

Change the criteria and you get a different, equally valid list. Start from Best by task.

Structured for people and AI

Copy for AI exports the context of the view you are looking at as structured text you can paste into an assistant:

• identity of each product and its page;
• the market, currency and snapshot date;
• the criteria and the priorities in force;
• facts, grades and ranks where the view has them, Value, and the source URL.

It does not create a new recommendation and does not change any calculation: it carries the same numbers the page shows, with the caveats that come with them.

Methodology versions

Rules improve, new facts appear and categories develop. Each category's rubrics and Value formula carry a version, so a grade or rank can change without the physical product changing — and that change can be traced to the version that caused it.

Category details

This page is the shared contract. What is specific to a category — its capability inputs, its scenarios and how it ranks — is below.

Independence and corrections

• Affiliate links may earn a commission. They never change facts, grades, weights, Value or order, and no product is boosted because a link pays more.
• Nothing here is a price forecast, a guarantee of quality, an official rating or professional purchasing advice.
• Used and refurbished offers are separate conditions with their own uncertainty: warranty, seller, returns, remaining lifetime.
• Found a wrong price, specification, link or match? Write to contact@thecomparator.tech — a corrected fact recalculates every number that depends on it.