What an email health score is, and what it is actually measuring

An email health score is a single number a vendor computes from several real measurements. It is not a standard, it is not reported by mailbox providers, and two tools will give the same domain different scores without either being wrong. Treat it as a dashboard light: useful for noticing that something changed, useless as a diagnosis. The inputs are what you act on.

What goes into a health score

Almost every implementation draws from the same pool. A reasonable set of inputs and weights, which you can compute yourself:

Two things to notice. Authentication and blocklist status are usually gates rather than contributors — fail either and the score collapses regardless of everything else. And a vendor with no seed network cannot measure placement at all, so its "health score" is inferred from the cheap signals only. That is the single biggest reason two scores disagree.

What a score cannot tell you

• Whether the problem is your list or your infrastructure. A clean infrastructure score with collapsing engagement means your list is the problem, and a composite hides that.
• Which provider is unhappy. Gmail, Microsoft and Yahoo disagree constantly. A blended number averages away the only actionable fact.
• Direction of causation. Low engagement lowers placement, and low placement lowers engagement. The score shows the loop, not its entry point.
• Anything about consent. A perfectly healthy domain can be sending mail it has no right to send.

Also note what it is *not*: Validity's Sender Score is a specific 0–100 IP reputation measure, not a health score, and Google retired its domain reputation chart in Postmaster Tools — so a tool still claiming to report "Google domain reputation" is inferring it.

Compute your own in fifteen minutes

Score each input yourself and keep the components visible rather than summing them:

dig txt example.com +short                       # one SPF record, <=10 lookups
dig txt selector1._domainkey.example.com +short  # DKIM key published
dig txt _dmarc.example.com +short                # policy + rua
dig -x 203.0.113.7 +short                        # PTR, must resolve forward

Then read a delivered message's Authentication-Results for spf=pass, dkim=pass, dmarc=pass with header.from aligned. Then pull these five numbers from the last 30 days:

Google's published complaint threshold is 0.3%, which is the one externally-set number in that table. The rest are working thresholds, not standards.

Improve the inputs in this order

1. Authentication and alignment. Binary, fixable in an hour, and it invalidates everything downstream while broken. One SPF record inside the 10-lookup and 2-void-lookup limits, DKIM signing at the selector you actually publish, a DMARC policy with rua, and alignment on the visible From domain.
2. Blocklist listings. Find the cause, stop it, then request delisting. Delisting while the behaviour continues does not hold, and repeat listings are harder to clear. Check Spamhaus, Barracuda, SpamCop, Abusix and UCEPROTECT, plus URI lists for your link domains. Validity's Heatwave launched on 3 September 2026; SORBS shut down in June 2024 and should be removed from any tooling.
3. Bounce and suppression discipline. Remove hard bounces on the first permanent failure, retry 4xx on a backoff, and enforce the suppression list at send time across every tool that mails as your domain.
4. Complaint rate. Easy one-click unsubscribe (both List-Unsubscribe and List-Unsubscribe-Post headers, acted on within two days), a visible footer reason-for-receiving line, frequency caps, and segmentation.
5. Engagement. Sunset addresses with no engagement in twelve months after one re-engagement attempt. This lowers your volume and raises your placement.
6. Transport hygiene. PTR that resolves forward, an EHLO name matching it, outbound TLS, RFC 5322-valid headers.
7. Consistency. Steady daily volume beats bursts. Warm new domains and new mailboxes before campaigning, and ramp no faster than roughly 30–50% a day while reputation is young.

What not to do

• Do not chase a vendor's number. Fix an input and re-measure the input.
• Do not compare scores across tools, or over time after the tool changes its model.
• Do not accept a score from a tool with no seed network as a placement measurement.
• Do not buy "reputation repair" services. There is no mechanism behind them.
• Do not treat a green score as permission to increase volume sharply.
• Do not test by mailing your own account; your mailbox has history with you.

MailPilot reports the inputs rather than a single opaque number: SPF, DKIM and DMARC verified against delivered mail, daily per-mailbox reputation tracking, and real inbox placement per provider from a seed-inbox network, alongside warm-up over OAuth or SMTP/IMAP.