How LaptopCompat verifies compatibility

LaptopCompat separates facts from conclusions.

A fact might be:

• a laptop contains two DDR5 SO-DIMM slots
• a USB-C port supports DisplayPort Alt Mode
• a laptop accepts USB Power Delivery input
• an SSD bay accepts M.2 2280 NVMe drives

Each fact is stored with its source.

Compatibility rules then compare laptop requirements with accessory specifications.

For example, we do not mark a USB-C charger compatible simply because its advertised wattage looks high enough. We also consider whether the laptop accepts USB Power Delivery and whether the charger supports appropriate output behavior.

When information is missing or contradictory, we prefer Unknown or Varies by configuration over pretending to know.

How we know what a page says

• Every fact on a model page has a source and the date we last checked it; both are shown on the page (which sources we trust, in what order).
• Every yes, no or likely comes from a written rule, not from a language model (how a verdict is reached).
• A number we calculated rather than read is labelled as an estimate (how estimates are labelled).
• Anything we cannot verify is shown as Unknown, never guessed. If something is wrong, tell us; we re-check sources on a fixed schedule.

1. Exact model identification

Laptop naming is messy — the same physical model can carry different marketing names, model numbers, machine types and regional SKUs. We keep meaningfully different variants (e.g. Intel vs. AMD configurations of the same generation) as separate records rather than merging them, since their memory, ports and display specs can differ.

2. Source hierarchy

When two sources disagree, the higher one wins:

1. The manufacturer’s own specification page for the exact model.
2. A manufacturer PDF: product brochure, user or owner’s manual, maintenance or service manual, or a quick-specs sheet.
3. Secondary sources, used to corroborate or where the manufacturer is silent, and always labelled as such: the chip maker’s own database for processor platform limits (for example the number of displays a processor supports), standards bodies and certification databases, and reputable technical reviews. For a few older chips a public encyclopaedia is cited for specification facts only.

Retailer listings and community reports (forums, user submissions) are recorded as what they are and never override a manufacturer document. Each source is stored with its type (manufacturer spec, manufacturer manual, manufacturer support, certification database, standards body, retailer, reputable review, user submission, or derived by us), so the type is visible next to every fact in our data.

3. Structured facts

Every fact used in a compatibility answer is stored with its source, retrieval date, and an extraction method (parsed automatically, AI-assisted with human review, entered by hand, imported via API, or derived by a rule).

4. Deterministic rules

Compatibility results come from rules written and tested in code — not from a language model guessing based on general knowledge. AI may assist with extracting or summarizing source material for human review, but it does not decide what is compatible.

5. Confidence levels

Results are labeled Verified, High, Medium, Low or Unknown, reflecting how directly the evidence supports the conclusion. We avoid presenting Low-confidence results as a purchase recommendation.

6. Conflict handling

When sources disagree, we prefer the newer authoritative manufacturer document where clearly applicable, check for SKU/variant differences, and preserve both facts internally rather than silently picking one. The public result becomes “Varies by configuration” or a note to verify your exact SKU, rather than false certainty.

7. How often we re-check

Models we list as current are re-checked when their last verification is older than about 90 days; discontinued models when it is older than about 365 days. A job runs every day at 06:00 UTC and re-fetches the cited sources of every model past its limit; another runs at 05:00 UTC and flags any source link that has stopped working. If a source page has changed, the model stays flagged as stale until a person has read the change and updated the facts, so a page can never silently change a number. A source that fails to load is flagged the same day. Accessory compatibility results are recomputed every Monday, and game and software verdicts are recomputed after every data update.

8. Corrections

Anyone can report a suspected error with the correction form, which every model page links to. Each report is stored with the page it concerns and is read by a person; a report goes from new to investigating, then is accepted, rejected or marked corrected. Nothing is published automatically from a submission. When a report leads to a change, the model appears in the “Recently corrected” list on the corrections page.

9. How “Can it run this game?” is judged

For games we compare the laptop’s graphics chip with the chip the publisher names in its requirements. Chips are placed in one of eight capability classes (tiers) by written rules: the manufacturer’s own class naming, generation, laptop-versus-desktop variant, and the publishers’ own “this or that” equivalences. A tier is a class label, not a benchmark score, and no frame rate is promised.

Meets needs the laptop to be clearly above the requirement. The same tier is only ever Likely, never a pass. If a laptop is sold with several graphics options, every option is checked: all meet it, it meets the requirement; none do, it does not; a mix is shown as Likely with each configuration explained, and you can pick yours. A stated video-memory floor caps a pass at Likely when the laptop has less video memory or only shared memory. A game that needs another operating system, or more soldered RAM than the laptop can ever have, is a definite no. Storage is shown for information only.

Known limits: tiers are judgments, not measurements, and can be off by one near a boundary; cross-vendor comparisons are the least reliable; Apple graphics are placed by chip class and core count; ray tracing, upscaling, drivers, cooling, resolution and frame rate are not modelled; publishers sometimes list optimistic alternatives; some older chips cite Wikipedia as a secondary source for spec facts only. Where we cannot place a chip or do not know a configuration, the answer is “Not judged yet”, which is never a “no”.

10. How estimates are labelled

We calculate a number only when the manufacturer states none and both inputs are sourced. The one estimate we make is the number of external displays a laptop can drive: the smaller of (the processor platform’s maximum number of displays, read from the chip maker’s own page, minus one for the laptop’s own screen) and (the number of video-capable outputs the manufacturer lists: HDMI, and USB-C ports with DisplayPort Alt Mode, Thunderbolt or USB4). It is shown as “Up to N (estimated)” with the note “Estimated from processor and ports — check your exact configuration”, and its sources are listed. A count the manufacturer states is never replaced by an estimate. We do not estimate frame rates, battery life, prices or anything else.