Is it your Wi-Fi, or is it past your router?

The problem

You already know the Wi-Fi is bad. You need to know why.

A call breaks up. A page hangs while somebody else is downloading. The line looks fine when you check it, so you restart the router and hope.

Every rival shows a device list and a speed test and never gives the answer. Perch gives the answer: the channel to switch to, and whether the fault is your Wi-Fi or sits past your router. It says so instead of guessing when it could not measure something.

The pair

Two apps, one core. The Mac does the diagnosis.

They are separate build products because their distribution is incompatible. Between them they cover the two places you notice bad Wi-Fi: at the desk, and in the room where the signal dies.

The two halves are not equal. iOS gives a normal App Store app no signal strength, no neighbouring networks, no channel and no band, and it hands back a spoofed hardware address, so the phone can never name the makers of the devices on your network. Perch does not work around that platform limit. The Mac does the diagnosis instead.

What it measures

Five checks. All five ship, all five are free.

Each one answers a question you can act on. The channel clause is the product: reading the air without naming the fix is what the free tools already do.

The watch

Leave it running and it keeps score.

Perch samples your link every 30 seconds for as long as you leave it running, draws the timeline as the samples land, and gives a verdict when the session ends. That verdict survives quitting the app, and a watch that ended because the battery ran low says so.

A stretch with no readings draws as a gap, never as a line through it. A line through a four-hour hole looks exactly like a healthy four hours, which is how a monitoring chart lies to you.

Free gets you one session, replaced by the next. Pro keeps 90 days. Alerts arrive while Perch is running and watching. There is no scheduling and no background service.

Honest limits

What Perch cannot do, in its own words.

The limits are the product. A tool that hides where its measurement stops will hide the answer too.

• The Mac has to be awake, and the lid has to be open. Perch holds the power assertion that blocks idle sleep, and that assertion has no effect on clamshell sleep, so a closed lid ends a watch. A Mac with no Wi-Fi link cannot watch; one on Ethernet with Wi-Fi still joined watches fine.
• Measuring costs you something. A Wi-Fi scan briefly interrupts the link it is measuring, so Perch says so under the result rather than letting you wonder why a call stuttered.
• The device list is one subnet, and incomplete on purpose. Other subnets, VLANs, IPv6-only devices and anything that ignores a probe stay invisible, and the app says so on the screen and in the export.
• Device identity decays. iOS 18 rotates a private Wi-Fi address about every two weeks on open or weak-security networks, so a device you labelled last month can reappear as a stranger. Perch marks those rather than guessing at a merge.
• Perch cannot always tell you how full your channel is. Channel utilisation comes from an optional beacon field most access points never broadcast. Perch shows it as a bonus when an access point volunteers it, never as the headline.
• The iPhone cannot see radio at all. No signal strength, no neighbouring networks, no channel, no band. That is iOS, not a choice Perch made.
• Attribution stops at your router. A powerline backhaul, a second router behind the first and a bloated modem your provider supplied all sit past the first hop and inside your own house. Perch does not name your provider's line.
• The thresholds are provisional. Most grading bands are labelled measured, conventional or provisional rather than presented as settled, and the labels change as the evidence does.
• "Could not measure" is an answer. Perch carries a fourth outcome for it instead of forcing every check into pass, warn or fail.
• Pairing rests on you comparing six digits. Both screens show the same number and you confirm they match. It is a human check, chosen over a heavier protocol on purpose, and the residual risk is written down rather than hidden.

Private by design

Your data stays on your network.

The Mac is the compute. No server of ours holds anything, because no such server exists.

• Scan results, device addresses, network names, IP addresses and hostnames go into a database file on your own Mac, at ~/Library/Application Support/Perch/Perch.store. It is named and namespaced on purpose, so you can find it and delete it.
• No server, no cloud, no account, no backend. Nothing to create, and nothing to leak.
• Once you pair, your own iPhone reads some of it across your own network, over TLS pinned to the one certificate you approved by comparing six digits on both screens. Network names and BSSIDs stay on the Mac.
• A Mac nobody has paired with, and nobody is pairing with, opens no port at all.
• Perch ships no analytics and no crash reporting. Neither app links an SDK for either one.
• Perch does reach the internet twice, and says so before it does: the connection check, and the bufferbloat download from Cloudflare. Both are named in full on the privacy page.

Free and Pro

The whole diagnosis is free.

The verdict and the evidence behind it cost nothing, on both platforms; the Mac's channel recommendation costs nothing too. That leaves Pro to sell the second question rather than the first: is it getting worse, and when.

Nothing else is gated. Every check that ships declares itself free in its own source file, and export gates on nothing.

Both apps will be free to download. Pro is planned as a one-time lifetime purchase alongside a subscription. Nobody can buy it yet, the purchase path is unbuilt, and there is no price to print.