How EZOBUY matching works

A small shortlist, with visible trade-offs and evidence limits.

1. Respect hard requirements

Budget, minimum installed RAM, required operating system, software platform compatibility and offer eligibility filter the catalog. Upgradeable RAM is not treated as RAM already installed. A chosen x86 architecture requirement excludes all non-x86 configurations, including ARM-based Macs. CUDA workloads require NVIDIA. Quarantined configurations are excluded.

2. Estimate fit

Your main use, battery and portability settings, and the visible extra-priority field determine dimension weights. Software compatibility and workload hardware gaps apply explicit penalties. CPU and GPU benchmarks are normalized only where an identity is mapped. Other dimensions may rely on proxies; these have lower evidence quality.

3. Account for uncertainty

Evidence confidence and authority affect ordering separately from the raw Match score. Low evidence can make a high-scoring device provisional. Offers older than 30 UTC calendar days, or with invalid, missing or future dates, are excluded from normal matching. Research mode can compare old or unavailable active records, but quarantined records stay excluded and stale handoffs stay disabled. The interface shows observation dates.

4. Test a close decision

The engine runs 200 deterministic variations of weights and evidence uncertainty. Close or fragile rankings can trigger one clarification. A selected answer changes an explicit preference or constraint and recalculates the shortlist.

5. Explain the limits

The catalog is curated, not exhaustive. Battery results are compared only within stored test-protocol cohorts; family-level test evidence is not represented as an exact configuration measurement. Match percentages describe modeled fit, not measured reliability or performance.

Public Beta RAM coverage

The catalog contains two 32 GB configurations and one 64 GB configuration. The 64 GB option is selectable. Higher-memory offers can exceed your budget, and missing coverage is shown rather than filled with invented configurations or upgrade assumptions.

Open “Evidence & methodology” in the finder for product sources, confidence, workload penalties and ranking stability.