One failed message out of two produces a 50 percent failure rate. That number is mathematically correct but says little about a domain's overall reliability. A low-volume domain needs a reporting method that keeps the underlying counts visible and separates a lack of observations from a demonstrated authentication result.
Aggregate reports describe the messages observed by participating receivers during stated intervals. They are not a complete census of every message your business attempted to send. RFC 9990 §3.1 defines aggregate report content and RFC 9989 §5.3.8 describes reporting participation limits; the investigation method below is an operational recommendation built around those limits.
Start with the domain's actual job
Ask whether the domain sends staff mail, occasional invoices, annual notices or nothing at all. Record its intended applications, owners and expected periods of use. A dormant brand that sends one renewal reminder each year is different from a parked domain that should never originate mail.
A quiet dashboard cannot resolve that distinction. Check the application inventory and business owner before changing policy or declaring a domain unused. Keep the planned future sender in the inventory even when it has no recent samples.
Put counts beside every percentage
References: RFC 9990 §3.1.1.8, RFC 9989 §5.3.8.
Use the aggregate record's message count, not the number of XML files or rows, when calculating observed volume. Display the reporting interval and participating receiver organizations beside the pass and fail counts. Explain whether duplicate reports, late arrivals and incomplete periods have been handled.
Illustrative case: Monday contains two observed messages, one failing. Tuesday contains 100, all passing. The combined observed failure rate is 1/102, about 0.98 percent. Averaging 50 percent and zero would produce 25 percent and misrepresent the message-weighted result. Neither calculation establishes outcomes at receivers that did not report. These numbers are synthetic, not a measured EmailAuth result.
Extend observation without inventing certainty
A longer window can capture an infrequent legitimate workflow. Choose it to include the relevant business event, such as month-end billing, rather than using an arbitrary number of days as proof. Mark policy or sending changes within that window so that distinct configurations are not silently blended.
A longer period still cannot reveal traffic outside reporting coverage. If no participating receiver observed the intended application, the honest conclusion is that this report set does not test that application. Do not describe zero reported failures as a statistically proven failure rate of zero across all mail.
Use controlled tests for missing routes
References: RFC 9989 §5.3.5, RFC 9989 §4.4.1, RFC 9989 §4.4.2.
Send a harmless test through the actual application to an authorized external mailbox. Record the visible From domain, envelope identity, DKIM signing domain, trusted receiver results and time. With an applicable valid policy and completed checks, a message-level pass requires a passing aligned SPF or DKIM path under DMARC; a standalone SPF pass under an unrelated return-path does not establish that result.
Test distinct routes separately. A manual message from a staff mailbox does not test an invoice service. Preserve the received-message evidence and distinguish it from aggregate observations. A controlled test establishes what happened to that test at that receiver, not universal delivery.
Decide with a coverage statement
Write a short conclusion naming what was observed, what was deliberately tested, what remains unobserved and who owns the next verification. Prioritize a single failure on a critical invoice route over the visual size of its percentage. Conversely, investigate unknown traffic without assuming every unfamiliar address is an authorized sender needing an SPF addition.
If a domain should not send, confirm that ownership decision and maintain a suitable non-sending-domain configuration through the normal change process. If it sends infrequently, preserve the test plan for its next event. In both cases, a useful report explains the evidence boundary instead of turning sparse data into a confident green or red status.
Keep the decision bounded
The local calculation gives 1 / 102 × 100 ≈ 0.9804% combined observed failure, versus 25% from an unjustified unweighted average of 50% and 0%. The ratio describes these supplied counts only. Test the actual business route separately and retain any unobserved recipients as a limitation.