InputProbe
Input device testers that publish their own measured error
A few days ago a link-building platform sent me a shortlist for inputprobe.com: ten gaming and tech sites, DR 55 to 80, €92 to €631 each, dofollow, most of them "permanent". The platform's senior SEO strategist had reviewed the list and called the fit excellent. The total came to about €2,700.
My product measures things for a living. inputprobe is a set of browser tools for mice, keyboards and controllers that publish how accurate each reading actually is, so checking a claim before trusting it is the day job. I gave each site about ten minutes. Here is what came back.
Three were real publications. A PlayStation news site with named editors, daily reviews, and a news-tip address kept separate from its advertising address. A print gaming magazine with published editorial and ethics policies. A streaming-analytics company whose reporters publish sourced, same-day stories. None of them sells guest posts, which is exactly why they are worth something, and exactly why no marketplace can sell you a link on them that survives.
One was a legacy tech brand that was sold years ago and now mostly rewrites Apple rumours and "$4 off AirPods" deals.
One was my direct competitor. It runs the same kind of browser controller tests I do, and its blog is already publishing articles aimed at my keywords. It was listed at €501 as a "perfect topical match". It is. That's the problem.
The other five were link farms, and none of them was subtle once I knew where to look.
Three "different" sites had byte-identical robots.txt files and the same Cloudflare nameserver pair. One of them had a guest-post page telling submitters to mention a fourth domain name in their email. They had copied the page from another site in the network and never changed it. The same page said, twice, that free guest posts are not accepted, listed "great backlinks" as the first benefit, and then accidentally published the operator's own list of Google search operators for finding guest-post sites.
Four of the ten block AhrefsBot in robots.txt while letting every other crawler in. That hides their outbound links from the one tool a buyer would use to see how many paid links each page already carries.
The "DR 80, 86K traffic" gaming site, the cheapest on the list at €92, turned out to be a team-name generator from 2022. Its last thirty posts were CBD capsules, kratom, betting, Czech tax advice and junk-car removal. The WordPress API lists exactly one author account for some 1,400 posts, and not one comment has ever been published. It was cheap because the market has already priced it as dead. The DR number just hasn't caught up.
Another listed a street address that doesn't exist, had a "Gambling" category sitting next to Minecraft in its menu, and credited its articles to a writer whose name looks generated. It had even published a profile piece about that writer.
Every money link I sampled on the five was dofollow with no rel="sponsored": casinos, CBD, and an Amazon SEO tool that got five links in a single "comparison" post.
If you get one of these offers, the checks that did the work:
- Read /robots.txt first. Blocking AhrefsBot specifically is a tell.
- Pick two sites from the same list and compare their robots.txt and - nameservers. Networks reuse both.
- On WordPress sites, open /wp-json/wp/v2/users. One account behind a thousand posts is a farm.
- Scroll the last thirty posts, not the homepage. The homepage is dressed; the archive isn't.
- Find the "write for us" page and read the small print. "No free posts" is a price list.
- Check whether the "perfect match" is your competitor.
I didn't buy any of it. The one useful thing to come out of the exercise was an editor's news-tip address at a site that doesn't sell links at all, which is the only kind of link I actually want.
If you've been sent a list like this and want a second pair of eyes on a specific site, ask in the comments and I'll say what I'd check.
Took the methodology out into the world this month: I posted the placement-error model to RTINGS’ test-methodology board, using their CPI article as the reference point. The basic result is that repeatability alone can’t catch a fixed offset because the same bias gets repeated every time.
Reading through their current methodology also taught me something: their actuator rig already publishes per-run spread. That changed the question I wanted to ask from “would you publish the spread?” to “how is the actuator’s traverse calibrated?”
That’s really the kind of conversation I want these tools to enable: methods criticizing methods, in public.
Next up: launch prep, then open-sourcing the polling estimator after launch.
Most online mouse testers measure your monitor, not your mouse: they count one movement per drawn frame, so a 1000 Hz mouse reads 60 on a 60 Hz display. I found this out by testing my own mouse and getting confidently wrong answers, and it annoyed me enough to build inputprobe.
The premise: browser hardware testers that behave like measurement instruments. Every tool reads raw device events instead of counting frames, publishes the accuracy band it was actually measured to have (results on A4 paper repeat within ~3%, on a bank card within ~6%), and refuses to print a number when the capture can't support one. The DPI test needs no ruler, it measures against objects exact by international standard; every bank card on earth is 85.60 mm long.
It's free: no accounts, no ads, nothing uploaded, everything client-side. The goal right now isn't revenue, it's earning the first 20–40 referring domains in six months against incumbents that rank on domain age with tools that measure the wrong thing.
Ask: if you've grown a free tool against entrenched SEO incumbents, what actually earned you your first real backlinks?
-
1
The unusual part is that the product is making uncertainty visible in a category where most tools appear to compete by presenting a single confident number.
That choice seems likely to shape how users perceive the product, especially when the market already has established answers people are accustomed to trusting.
-
1
You've put your finger on the exact bet. A single confident number is the better demo, it feels authoritative, it screenshots well, and nobody asks follow-up questions. The problem in this category is that the confident number is often measuring the wrong thing entirely (usually the monitor's refresh rate), and the user has no way to tell from the outside. Confidence and correctness are unrelated here, which is the worst possible version of that trade.
The wager is that the audience splits. Casual users may genuinely prefer the tool that says "800 DPI" full stop, I'll lose some of them and I've accepted that. But most people who go looking for a DPI test are there because something already feels wrong, their aim, a new mouse, a driver update. They've been burned by a confident answer once. For them, "548 DPI, three passes within 1.3%" answers the question they actually have, which is "can I act on this?"
Two design choices try to soften the perception cost: the median still leads (big number first, band second, so it reads as a result rather than a shrug) and when the tool refuses to print a number it always says why and what to change, so uncertainty arrives as instructions instead of doubt.
The precedent that gives me some hope: weather apps normalized "70% chance of rain" and nobody reads that as weakness. Categories can learn to treat visible uncertainty as competence, usually right after enough people get burned by false confidence.
Out of curiosity, when you first saw the error band, did you read it as credibility or as hedging? That first-impression read is the one thing I can't measure from inside.
-
1
That’s helpful context. I appreciate you unpacking the thinking behind it.
I’d like to continue this conversation outside the thread. What’s the best email to reach you on?
-
1
Glad it was useful. Happy to, you can reach me at contact[at]inputprobe. What did you have in mind? Asking mostly so I can point us at the right thing, and if any of it is useful to other people building in public, I'm equally glad to keep it here in the thread.
-
1
Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.
-
-
-
-
Comment