Research & testing

Our testing methodology

Our research process and proposed test protocol: separate detection, removal, false positives, and performance, with evidence and limits before publishing results.

No independent malware detection, removal, false-positive, or security-product performance tests have been performed for this project. Tool profiles summarize official documentation. The website’s build, link, layout, accessibility, and search checks verify the website; they do not measure the effectiveness of the security tools it describes.

This page defines our current research process and proposed version 1 requirements for future product tests on Windows or Mac. It contains no test results. A future test would need its own documented plan, qualified execution, evidence, and review before receiving a tested label.

What the evidence labels mean

Readers should be able to distinguish a documented feature from something independently observed. A vendor’s statement that a tool supports a task can justify investigating that tool’s suitability; it cannot supply our own effectiveness score.

Evidence typeWhat the site can claimWhat it cannot establish
Official documentation researchThe documented feature, scope, requirement, or procedure at the checked dateThat we exercised it or measured its success
Independent interface checksThe exact controls and behavior observed in a recorded version and environmentDetection or cleanup of threats not included in the check
A published threat experimentThe observed outcome for defined cases under the recorded protocolUniversal protection or performance on every device
An unverified reportThat a claim needs investigation, clearly attributedA tested fact, rating, or recommendation based on measured success

The current tool profiles carry researched status. A future hands-on claim must point to the corresponding record and describe its scope. If only installation or an interface was checked, the article must say so rather than implying malware efficacy. Writing completion, search indexing, and human technical approval remain separate states.

How documentation research is performed

For product features and procedures, start with the vendor’s current support or technical documentation. Record the source, supported platform, relevant version or entitlement, and verification date. Check whether the source concerns the same product: Windows Defender Antivirus, for example, is not interchangeable with every product using the Defender name.

Compare related documents when a feature depends on a free tier, paid plan, architecture, or managed configuration. If official pages disagree, show the dated discrepancy and the basis used rather than inventing certainty. The Malwarebytes profile illustrates a documented feature-table and scan-article discrepancy. Documentation research does not resolve that conflict through product execution.

Write the reader’s decision path and then perform a separate editing pass for scope, action consequences, ordinary alternatives, and useful links. Keep vendor claims attributed. Source checking dates show when we examined the material; they do not make a changing product permanently current.

Define the question before collecting results

A future plan must specify whether it tests prevention, on-demand detection, removal of an established condition, false positives, or performance. These questions need different starting states and outcomes. Finding a file during a manual scan cannot be reported as blocking its execution in real time.

Define the product version, plan, configuration, operating-system build, hardware or virtual environment, update policy, connectivity, user actions, and completion criteria. Set the treatment of failed setup, crashes, interrupted cases, and unavailable cloud services in advance. Publish deviations alongside the results rather than quietly selecting successful runs.

AMTSO’s Testing Protocol Standard version 1.3 provides a reference for transparent plans, product configuration, sample selection, and evidence. Our proposed requirements draw on testing guidance but are not a claim of AMTSO compliance, certification, or endorsement. Any future standards claim would require the relevant process and confirmation.

Establish the environment and case provenance

Use an appropriate isolated research environment under qualified supervision. A future test must document containment, permitted connectivity, access restrictions, reset procedure, and checks that each case starts from the intended state. Personal accounts, production devices, and unrelated networks must be excluded from the experiment. Live-threat work requires a separate qualified research plan; these requirements are not a home testing procedure.

For each case, retain an identifier, acquisition and validation dates, provenance, relevant behavior, and classification rationale. State the selection window, deduplication rules, exclusions, and why the cases fit the tested use case. Several files from one incident can represent one installation; the counting unit must be explicit.

Adware and PUP classification needs particular care. An unwanted-software finding or preinstalled-utility listing is not automatically a confirmed malicious sample. Document the behavior and vendor criteria, with disputed cases separated from established malware. See the PUP investigation for that distinction.

Retain enough restricted evidence for qualified review without distributing malware or exposing personal information. The public report should describe provenance and validation, with sensitive details redacted and those redactions explained. Any future sample handling or third-party submission needs its own authorization and disclosure policy.

Separate detection from completed remediation

Record the stage at which the product acted: download, access, execution, manual scan, or later behavior. Preserve the exact finding and action. A blocked installer, quarantined file, and removed installed component are different outcomes.

For a removal experiment, establish and document the condition before cleanup, then record selected actions, restart requirements, remaining components, and the original behavior afterward. A second scan finding nothing is one observation; it is not sufficient by itself to establish that persistence, browser configuration, or related exposure was resolved.

Keep incomplete actions, failures, and recurrence visible. Separate prevention outcomes from removal outcomes rather than combining them into one convenient success figure. Local remediation also cannot undo information already stolen, so conclusions must match what the experiment actually measured.

Evaluate false positives with a separate benign set

AMTSO’s archived false-positive guidance emphasizes legitimate content, context, and validation. A future benign set must have a defensible source and status, with ordinary filenames, locations, and intended use. A familiar publisher name alone is not ground truth.

Record whether a legitimate item was warned about, blocked, quarantined, or made unusable. Describe the consequence for its actual task. Keep PUP policy decisions and inconclusive classifications separate from confirmed false positives, and document how disputes were resolved. The false-positive glossary explains why disagreement alone does not settle a finding.

Report the eligible benign cases and the observed events together. A small or narrowly selected clean set cannot support a general promise of zero false positives.

Measure performance against a defined baseline

AMTSO’s archived performance guidance is a reference for designing measurements around user tasks and controlled conditions. A future plan must identify the baseline, workload, hardware, power state, background activity, product settings, and repetitions.

Separate installation, update, first scan, later scans, and ordinary application tasks. Report elapsed task time and relevant resource observations with units, run counts, and variation. State how warm-up, caching, and run order were handled. One fast scan or one Task Manager snapshot is not a general measurement of device impact.

Comparisons must use equivalent intended tasks and disclose differences in scope. A focused AdwCleaner cleanup and a broader antivirus scan can inspect different content; their durations cannot establish a universal speed winner without that context. Virtual-machine measurements must be labeled as such rather than presented as the experience of every physical computer.

Keep demonstration checks in their proper scope

EICAR’s explanation describes a non-viral file designed to trigger a recognizable antivirus response. AMTSO’s test-file guidance addresses the limits of such checks. Observing a demonstration response can help check that a particular handling path works; it cannot represent a collection of real threats or an established infection’s cleanup.

The site has not performed even that demonstration check. No test-file contents, executable downloads, or malware samples are provided here. A future report must label a demonstration separately and avoid converting it into a detection percentage or effectiveness ranking.

Publish only conclusions the record supports

A future result needs the dated plan and methodology version, environment and product details, case accounting, observed outcomes, retained evidence, deviations, and limitations. Publish the numerator and denominator for any proportion, identify excluded or inconclusive cases, and explain uncertainty. Closely related samples and small collections limit broader conclusions.

Disclose product acquisition, funding, commercial relationships, and any vendor involvement. Technical feedback must be evaluated against evidence, and commercial participation must not determine sample selection, rankings, or suppressed failures. The editorial policy and affiliate disclosure describe the site’s corresponding boundaries.

Check the design, supporting evidence and conclusions before publishing a tested label or result. Correct material errors through the corrections policy, retaining the reason and affected version. Reusing a result after a product or methodology change requires an explicit relevance check; an old observation does not automatically describe a new release.

This methodology was researched and edited on 6 October 2026. The cited protocol is dated 2019 and the specialist guidance includes archival documents; their principles inform planning, while current product requirements need separate verification. There are no independent detection rates, removal results, performance benchmarks, or lab rankings to report.

Sources & documentation

Sources checked on 6 Oct 2026. Documentation research does not establish independent testing.

MalwareCleanupTools

Guides based on official documentation. About our content · Corrections policy