How We Test

Last updated: June 14, 2026

Every product recommendation on Axis Intelligence comes from real testing. This page documents how each member of our editorial team tests the products they cover — what equipment we use, how long we use it, and how we form conclusions.

If you want to know the editorial principles behind these methodologies — our independence, our funding, our correction policy — read the Editorial Policy.

General principles

Across every category at Axis Intelligence:

• No product is recommended without hands-on use. If we have not used it, it does not get a recommendation.
• Testing happens in real-world conditions. Homes, offices, real networks, real workloads — not lab benches with synthetic test suites alone.
• Minimum testing windows are enforced. The shortest test cycle is seven days. Most reviews are based on 30+ days of use.
• Methodology is disclosed per article. Every review article includes a “How we tested” section calling out what hardware, software, environment, and time period the conclusions are based on.
• We retest at major updates. When a product receives a meaningful update — new firmware, a hardware refresh, a new pricing tier — we re-evaluate the affected reviews and best-of articles.

Cybersecurity, VPN, and Privacy Tools

Tested by Marcus Chen

What gets tested

VPNs, antivirus software, password managers, two-factor authentication apps, secure email services, encrypted messaging apps, identity theft protection services, and browser-based privacy tools.

How VPNs are tested

• Speed testing on three baseline connections (1 Gbps fiber, 250 Mbps cable, 50 Mbps DSL), measured at three geographic distances (same-country, transatlantic, opposite hemisphere) using Ookla Speedtest CLI, Cloudflare, and fast.com — minimum five runs per server, average reported.
• Streaming unblocking verified against Netflix, Hulu, BBC iPlayer, Disney+, HBO Max, Amazon Prime Video — six countries each, retested monthly.
• Leak testing via dnsleaktest.com, ipleak.net, browserleaks.com — DNS, WebRTC, IPv6, and browser fingerprint leaks all checked on Windows, macOS, iOS, Android, and Linux clients.
• Privacy claims verified by reading the privacy policy, jurisdiction, ownership, and any available independent audit reports (e.g., Cure53, Deloitte). No-logs claims are reported with the verification basis.
• App reliability observed over a 30-day daily-use period on at least three devices.

How antivirus and security suites are tested

• Detection assessed using real malware samples in isolated VMs (Windows 11 sandbox), supplemented by AV-Test and AV-Comparatives’ most recent independent benchmarks.
• System impact measured: boot time, file copy speeds, archive operations, browser launch — before and after installation.
• Real-world usability scored on 30 days of daily use: false positives, intrusiveness, update behavior, support quality when contacted as a regular customer.

How password managers are tested

• Setup tested on Windows, macOS, iOS, Android — full vault migration from at least one competitor.
• Browser integration tested on Chrome, Firefox, Safari, Brave.
• Recovery scenarios tested: lost master password, lost two-factor device, account compromise.
• Sharing, family vaults, and business team features tested with multiple test accounts.

What I don’t do

• I don’t recommend a VPN I haven’t personally subscribed to and used as my daily VPN for at least two weeks.
• I don’t repeat marketing claims about military-grade encryption, world-class privacy, or fastest network in the world without verifying what those claims actually mean.

Artificial Intelligence and AI Tools

Tested by Sarah Mitchell

What gets tested

Large language model chatbots (ChatGPT, Claude, Gemini, Perplexity, Copilot, and competitors), AI writing tools, AI image and video generators, AI coding assistants, AI productivity assistants, AI translation tools, and AI detection tools.

How AI chatbots are evaluated

• Standard task battery run against each model: code generation in three languages, document summarization, structured data extraction from messy text, multi-step reasoning, factual recall on niche topics, refusal handling for ambiguous requests.
• Real-world workflow integration tested by using the tool as a daily working assistant for at least two weeks — not just benchmark sessions.
• Hallucination rate tracked: when the model produces confident-sounding but verifiably wrong output, it’s logged.
• Pricing per real use case calculated: tokens, message limits, file upload caps, API costs for typical workloads.

How AI image and video tools are evaluated

• Identical prompts run across competing tools, outputs compared side by side.
• Iteration cost measured: how many tries are needed to get a publishable result.
• Style flexibility, prompt adherence, and handling of human anatomy, hands, and text-in-images explicitly tested.
• Content moderation behavior documented.

How AI productivity tools are evaluated

• Real workflow tested: meeting transcription tools used on actual meetings, AI writing assistants on actual articles, AI email assistants on actual inbox.
• ROI calculated against the same workflow done manually — time saved, quality difference, error rate.

What I don’t do

• I don’t write about an AI tool I haven’t used for at least a week.
• I don’t trust benchmark scores or vendor demos as a substitute for daily-use testing.
• I don’t claim a model is “better than ChatGPT” without showing the comparison.

Networking, Web Hosting, and Self-Hosted Tools

Tested by James Porter

What gets tested

Web hosting providers, VPS hosting, managed WordPress hosting, CDN services, domain registrars, mesh Wi-Fi systems, routers, network-attached storage, Linux distributions, self-hosted application stacks, Docker images, and home server hardware.

How hosting providers are tested

• Real WordPress site deployed on every reviewed provider — same theme, same plugins, same content — for at least 30 days.
• Performance measured with GTmetrix, PageSpeed Insights, and WebPageTest from three geographic locations, minimum 10 test runs per location.
• Uptime monitored with UptimeRobot for the full review period.
• Support quality tested by submitting at least three real support tickets (one billing, one technical setup question, one escalation) and timing responses.
• Migration tested both into and out of the host.
• Backup and restore tested end-to-end.

How routers and Wi-Fi are tested

• Deployed in a real residence (not a lab): three-story house, mixed device load (40+ devices), real wall materials.
• Coverage measured with NetSpot at fixed test points before and after installation.
• Throughput tested at 1 meter, 10 meters, and behind two walls using iPerf3 to a wired server.
• Mesh roaming behavior tested with a moving smartphone.
• Long-term reliability monitored for 30+ days: crash rate, firmware update behavior, app reliability.

How self-hosted tools are tested

• Installed on a real home server (Proxmox host with multiple LXC containers) and on a cloud VPS.
• Configured for real use, not just spun up.
• Update path tested across at least one minor version release.
• Backup and restore tested.

What I don’t do

• I don’t recommend a hosting provider I haven’t run a production-style site on.
• I don’t quote uptime figures from the host’s marketing — only what I measured myself.

SaaS, Productivity, and Business Tools

Tested by Elena Rodriguez

What gets tested

CRM systems, email marketing platforms, project management software, accounting and bookkeeping tools, HR and payroll software, marketing automation, customer support helpdesks, e-signature tools, scheduling software, and business survey tools.

How SaaS tools are tested

• Real account signed up at the actual pricing tier the average small business would use — not a special review tier.
• Real data imported. When I test a CRM, I import a real contact list (500+ records from a Test business). When I test accounting software, I run a full month of test transactions.
• Typical workflow run for 30 days. Send actual email campaigns, run actual projects, book actual demos.
• Renewal experience tracked. When a trial ends, what happens? Hidden fees? Difficulty downgrading? Auto-renewal pricing jumps?
• Support quality tested with realistic small-business questions submitted as a regular customer.

How CRM and email marketing tools are scored

• Setup time from zero to first usable workflow.
• Total cost at the size of an average small business (50 users for CRM, 5,000 contacts for email).
• Integration support with the tools small businesses already use (QuickBooks, Slack, Google Workspace, Shopify).
• Deliverability tested for email marketing platforms using established testing services (GlockApps, Mail-tester).

What I don’t do

• I don’t review a SaaS tool I’ve only seen demoed.
• I don’t trust a vendor’s case studies — I run my own test workflow.

Tech Careers, Salaries, and the Job Market

Researched and written by David Park

What gets covered

Salary benchmarks, certifications, the job market across tech roles, coding bootcamps, layoff trends, hiring practices, and career transition guides.

How I research salary data

• Primary sources only. Salary figures come from U.S. Bureau of Labor Statistics Occupational Outlook Handbook, Levels.fyi (for FAANG and similar), Stack Overflow Developer Survey, Robert Half, Dice Tech Salary Report, and PayScale.
• Date-stamped. Every salary number cites the source year.
• Geographic and experience breakdowns provided. A “software engineer salary” without breaking down by location, years of experience, and company size is not useful.
• Cross-validation. When sources disagree (Levels.fyi vs. BLS, for example), I note the disagreement and explain why.

How I research certifications and bootcamps

• I look at outcome data from Council on Integrity in Results Reporting (CIRR) for bootcamps that report through it.
• I track real graduate outcomes — job placement rate, salary, time to first job — not marketing claims.
• For certifications, I review the actual exam curriculum, retest cost, market employer demand on LinkedIn/Indeed, and post-cert salary differential.

How I research layoffs and job market trends

• Layoffs.fyi for the structured tracker.
• BLS JOLTS for hires/openings/separations.
• Indeed Hiring Lab and LinkedIn Economic Graph for posting trends.
• Challenger, Gray & Christmas for monthly layoff announcements.

What I don’t do

• I don’t recommend a bootcamp without checking its outcome data.
• I don’t cite a salary “average” without saying where it came from.
• I don’t claim the job market is great or terrible without citing the data.

Smartphones, Laptops, Tablets, and Wearables

Tested by Alex Rivera

What gets tested

Smartphones (iPhone, Android), laptops (consumer, business, gaming), tablets, smartwatches, fitness trackers, wireless earbuds, over-ear headphones, and the major mobile carrier plans.

How smartphones are tested

• Daily driver for at least 14 days. Every reviewed phone replaces my primary phone for two weeks minimum. Big reviews go 30+ days.
• Real battery life measured under a realistic mixed workload (calls, social media, navigation, photos, video streaming, gaming) — not synthetic battery tests.
• Camera tested in normal conditions. Indoor mixed light, outdoor daylight, low light at dusk, night mode comparisons, video in walking shots, portrait mode on real people. Sample shots are published, not just verdicts.
• Performance tested with actual apps, not just Geekbench: heavy game (real frame rates), video edit (real export times), large PDFs, multitasking with 10+ apps.
• Connectivity tested in known weak-signal locations: train, basement, distance from router.
• Heat and throttling measured during sustained workload.

How laptops are tested

• Daily driver for at least 14 days for major reviews.
• Battery measured during real work (browser tabs, video calls, light coding or document work), not synthetic loop tests.
• Display measured with a colorimeter (X-Rite i1Display Pro): brightness, contrast, gamut coverage.
• Speakers, keyboard, trackpad, webcam assessed subjectively over the test period.
• Thermals tested under sustained load.

How earbuds and headphones are tested

• 30-day daily use cycle minimum.
• Tested across genres of music with reference tracks.
• Active noise cancellation tested in three environments: airport-like ambient noise, coffee shop chatter, traffic.
• Call quality tested on both ends of a real call.
• Comfort assessed at 4+ hour wear sessions.

What I don’t do

• I don’t review a phone I’ve only had for a weekend.
• I don’t trust spec-sheet camera comparisons without sample photos.
• I don’t rate a laptop based on Geekbench alone.

Smart Home, Cameras, and Home Automation

Tested by Jeff Aaron

What gets tested

Smart doorbells, indoor and outdoor security cameras, smart locks, smart lighting, smart thermostats, smart plugs and switches, robot vacuums, home security systems, smart speakers, and Matter/Thread-enabled devices.

How smart home devices are tested

• Installed in a real home, not a controlled lab. Walls block Wi-Fi. Dogs trigger motion sensors. Family members complain about smart switches that act weird at 2 a.m.
• Minimum 30-day test period for every reviewed product.
• Integration tested with the major platforms: Apple HomeKit, Google Home, Amazon Alexa, Matter — including the ones the product claims to support that don’t always work as claimed.
• App reliability tested over the full period. Smart home reviews are often broken by apps that crash, lose connection, or push bad updates a month in.
• Power consumption noted for always-on devices.
• Privacy assessed: what data the device sends, to where, and what data the user can opt out of.

How security cameras are tested

• Deployed indoors and outdoors in real conditions (rain, glare, low light, full dark with IR).
• Real motion events from people, pets, cars, and “false positive” sources (shadows, branches in wind, headlights through windows) all logged.
• Video quality assessed at advertised resolution and the lower-bandwidth fallback resolution.
• Cloud storage costs calculated for typical motion event volumes.
• Local storage tested if available.

How robot vacuums are tested

• Same room, same dirt simulation (Cheerios, sand, hair on a test rug) across competing vacuums.
• Mapping accuracy assessed after multiple full runs.
• Obstacle handling tested with deliberately placed objects.
• App reliability and mapping persistence tested for 30+ days.

What I don’t do

• I don’t recommend a smart home device after a single weekend of testing.
• I don’t trust Matter compatibility logos until I’ve connected the device through Matter myself.

Gaming Hardware and Esports

Tested by Noah Mc

What gets tested

Gaming laptops, pre-built gaming PCs, GPUs, CPUs, mechanical keyboards, gaming mice, gaming headsets, gaming monitors, controllers, handheld gaming consoles (Steam Deck and competitors), and major game streaming platforms.

How GPUs are tested

• Tested on a controlled test bench (Ryzen 9 7950X, 32GB DDR5, NVMe SSD, latest drivers) for benchmark consistency.
• Real games at real settings, not just 3DMark. Cyberpunk 2077, Spider-Man 2, Counter-Strike 2, Valorant, Fortnite, Apex Legends, Black Myth: Wukong, and a rotating list of new releases.
• 1080p, 1440p, and 4K results all reported.
• DLSS, FSR, XeSS upscaling all tested where applicable, with image quality assessed not just frame rate.
• Power draw, fan noise, and thermal throttling measured.

How gaming laptops are tested

• Real games at the laptop’s native resolution and refresh rate.
• Battery life on gaming AND on regular productivity use (a gaming laptop is also a daily driver for most owners).
• Fan noise measured under sustained load.
• Thermal throttling and sustained performance after 30+ minutes.
• Display measured for response time and HDR claims actually verified.

How peripherals are tested

• Keyboards — used as my daily driver for at least two weeks. Typing, gaming, audio (sound test), key feel, lubing as shipped, switch consistency.
• Mice — 30-day daily use, sensor tested with paint test for tracking accuracy and angle snapping, click latency measured.
• Headsets — gaming sessions (positional audio, footstep tests), music listening, voice chat call quality from the receiving end.
• Monitors — colorimeter measurements (response time, color accuracy, gamut coverage, HDR claims), motion clarity tested with UFO test, real gaming and real productivity use.

What I don’t do

• I don’t quote frame rates from synthetic benchmark suites alone.
• I don’t recommend hardware I haven’t gamed on for at least a week.
• I don’t trust “esports tournament-grade” marketing without checking the actual specs.

Electric Vehicles, Solar, and Clean Energy

Tested by Aidan Jad

What gets covered

Electric vehicles (test drives where possible, owner experience as a daily Hyundai Ioniq 6 driver), home and public EV chargers, solar panels and installers, home battery storage, heat pumps, e-bikes, and the underlying economics of the clean energy transition.

How EVs are evaluated

• Daily driver experience for vehicles I have access to. As a Hyundai Ioniq 6 owner in Montreal, I’ve lived through what real cold-weather EV ownership looks like — winter range, public charging in extreme cold, snow tire impact on range, heated steering wheel as a battery saver.
• Range tested in real conditions, not just EPA estimates. Highway speeds, cold weather, hot weather, with a roof box, with the heat on full.
• Charging curves measured. Not just “DC fast charging up to X kW peak” but the actual time from 10% to 80% on a real network charger.
• Total cost of ownership calculated with real numbers: depreciation, charging cost at home and on the road, insurance, tires (EVs eat tires faster), maintenance.
• Software updates and infotainment tested through real ownership, not a press loan demo.

How home EV chargers are evaluated

• Installed on real residential service.
• Charging speed measured against the vehicle’s onboard charger limit.
• App reliability, scheduling, and time-of-use rate integration tested.
• Cold-weather performance (Quebec winter) noted.

How solar panels and installers are evaluated

• Per-kW installed cost benchmarked against the National Renewable Energy Laboratory’s benchmark report and the Solar Energy Industries Association data.
• ROI calculated with real local electricity rates, real net metering policies (which vary by state and province), and the federal tax credit (subject to political changes — current status noted at publication time).
• Installer reputation assessed using NABCEP certifications, BBB ratings, Solar Reviews, and local permitting and complaint data where available.
• Real installations consulted where possible.

How heat pumps and home batteries are evaluated

• TCO modeled at real local energy prices.
• Efficiency claims (HSPF for heat pumps) checked against AHRI directory.
• Cold-climate performance (relevant in Montreal) emphasized — many heat pumps lose efficiency drastically below -15°C.

What I don’t do

• I don’t quote EPA range as if it were real-world range.
• I don’t recommend a solar installer based on a paid sponsorship.
• I don’t repeat “EVs save you money” without doing the actual math for the reader’s situation.

Statistics and Data Reports

Researched and published by Axis Intelligence Research

What gets published

Industry statistics articles, data reports, market analyses, and original Axis Intelligence Research surveys and studies — across all of our coverage areas.

How statistics articles are sourced

• Primary sources only. We cite government data, academic research, named research institutes, vendor research reports, and reputable industry trackers. We do not aggregate from other content sites.
• Source hierarchy: government (.gov) → academic (.edu) → named research institutes (IBM, Verizon, Gartner, IDC, BloombergNEF, IEA, etc.) → tier-1 media that themselves cite primary sources.
• Date-stamped per statistic. When the source published the figure is shown alongside the figure.
• Methodology disclosed when relevant (e.g., a 1,000-person survey vs. comprehensive government census).
• Quarterly review. Every published statistics article is re-checked at least every three months; if the source has updated its data, the article is refreshed and the “Last updated” date changed.

How original Axis Intelligence Research is conducted

When we run our own primary research (surveys, tests, scraped public data):

• Methodology published in full alongside the report.
• Sample size disclosed for surveys.
• Limitations explicitly named. Self-selection bias, geographic limitations, response rate, anything that affects how the result should be interpreted.
• Raw data made available for download under CC BY 4.0 license where privacy allows.
• Independent review before publication when the topic warrants it.

How citations work

• Every statistics article carries a citation block (APA, MLA, Chicago) at the bottom for readers who want to cite our work.
• Our data is available for republication under CC BY 4.0 with attribution.

Updates to this page

This methodology page is reviewed every six months and updated as our testing practices evolve — new equipment, new test methods, new product categories. The “Last updated” date at the top reflects the most recent material change.

If you spot a gap in how we describe our methodology, or you think a category needs a more rigorous test, we want to hear about it: [email protected].