How we measure things
Every number on outsolvi.com is either a figure from a published source or a setting read from our own product. This page shows which, and how we check both, so you can decide for yourself whether to trust us.
Checking our work? The same standard applies to your own sends: every open graded on how likely it was a real person, with the reason shown. 14 days free, no card. Start free
How opens are graded
Outsolvi grades every open when it arrives. It looks at who fetched the tracking image and how: the address and network the request came from, the User-Agent, how soon after delivery it happened, and whether a reply or a click on the same email backs it up. Each open gets a label on the email's timeline, with the reason in plain words, such as "Microsoft Defender security scan" or "opened in Outlook desktop".
- Verified: a reply arrived on the thread, so a person read it. Counted.
- Most-likely human: the request looks like a person opening the email. That includes Gmail's and Yahoo's image proxies, which fetch the image when someone opens the message. Counted.
- Possibly human: too little to go on either way. Shown, not counted.
- Apple Mail: Apple Mail Privacy Protection fetched the image on Apple's servers, which happens whether or not anyone reads. Counted only if the same person clicks or replies.
- Machine preview and Link preview: an automated service fetched the image or one of your links, not a person. Not counted.
- Scanner: a security gateway such as Microsoft Defender, Mimecast or Proofpoint checked the email or its links. Not counted.
- Your own open: you opened your own email, for example in your Sent folder. Not counted.
Two details matter for Gmail. Repeat fetches through Gmail's proxy within ten minutes are grouped as one open, because one read can make several fetches. And if you send from a gmail.com address, a fetch in the first five minutes is marked as possibly your own Sent folder, because opening your Sent folder goes through the same proxy.
How we measure whether the grading is right
There is no held-out test set and no single accuracy percentage. We are not going to quote one, because we would have to make it up.
What runs instead is a weekly job. For every rule that fired at least five times in the previous thirty days, it computes one number: of the events where a user told us something, how many confirmed the rule against how many overrode it. A confirmation is a reply arriving on a thread we graded as human, or a click strengthening an open. An override is you pressing Not me on your own open, or telling us an event we filtered really was the recipient.
The result is stored per rule, so a rule that starts misfiring shows up as its own number falling, rather than being averaged into a headline figure that stays reassuring. When a rule slips, our team changes it. Nothing retrains itself.
A rule without enough feedback reports nothing at all. Not a default, not an estimate, not last quarter's number.
The honest limitation: this measures the rules against the corrections people bother to make. Someone who never presses Not me is invisible to it. That is a real signal from real usage rather than a laboratory score, and the two are not the same thing.
The scores behind the labels
| Grade | Score | Counted as an open? |
|---|---|---|
| Verified | 100 | Yes |
| Most-likely human | 70 | Yes |
| Possibly human | 20 | No, below the floor |
| Machines and your own opens | 0 | No |
The counting floor is a score of 25. An event at or above it updates your open count; anything below is stored with the reason it was set aside and never counted. The floor was lowered from 40 to 25 in March 2026, on the view that counting a borderline real read is a smaller error than silently dropping one.
These are discrete scores, not a continuous 0-100 gauge. The engine assigns one of the four values above, so an event never scores 63 or 87. In the app you see the label and its reason, not the score.
Where our numbers come from
We don't publish statistics about our customers' email. Every other number on this site is one of two things: a figure from a published study or report, listed with a link on our research and sources page, or a setting read from our own product, like the scores above.
If we can't find a source for a number, it doesn't go on the site. In September 2026 we checked the figures on outsolvi.com against that rule and took out the ones that failed it, including open-rate multipliers and follow-up conversion rates that had no study behind them. The pages that leaned on them most carry a note saying so.
When we quote research, we name who ran it and when, and we say where it stops. The best-known research on response speed, for example, measured how fast companies called back inbound web leads, not email threads, and our follow-up pages say that.
Some of our free tools estimate things, such as how much Apple Mail pre-fetching pads an open rate. Their starting values are guesses, labeled as guesses, and you can change them.
How we check what the site says
Claims about what Outsolvi does are checked against the product itself: the web app, the Outlook add-ins and the Gmail extension. That's how we know, for example, that classic Outlook needs the desktop add-in and that Outlook for Mac doesn't track yet, which is why the site says so.
When the product changes, the pages that describe it change with it. When we find a page that describes something the product doesn't do, we fix the page rather than leave it for later.
When pages get updated
Long-form pages show the date they were last updated. The date moves when the content changes, not on a timer.
We re-check a page when something it depends on changes: a product release, a pricing change, a new study, or a mailbox provider changing how it handles images and links. Comparison pages get re-checked when a competitor changes its product or prices and we learn of it.
If you find an error or an out-of-date claim on any page, email us at support@outsolvi.com and we'll fix it.
Recent changes to the grading
Updated September 2026
- Every open shows its reason. Open a tracked email in Outsolvi and the timeline shows the label and the plain-words reason for each open and click, counted or not. The honest version of "most accurate" is "check for yourself."
- The rules are checked every week. Mailbox providers and security scanners change their behavior without notice, so every rule's precision is recomputed weekly from opens a reply later confirmed and from user corrections, and our team changes a rule when it slips.
- Gmail and Yahoo proxy opens count. Their image proxies fetch the image when someone opens the email, so a proxy open is graded most-likely human. Apple Mail Privacy Protection pre-fetches stay out of the count unless the reader clicks or replies.
See it on your own inbox, not ours.
Start Free TrialSee it on your own inbox
14-day free trial. Every open graded from day one.