Back to blog
Threat Intel
Phishing Forensics

DMARC pct Tag Realities: Google, M365, and SEG Discrepancies

Setting p=reject with a low percentage looks like a safe rollout strategy until you realize major receivers handle the math completely differently.

MailSleuth Research
Email Security Team
October 27, 20269 min read
Abstract illustration of an email pipeline splitting into three different filtering mechanisms.

You just flipped a massive enterprise domain to a strict reject policy, but out of an abundance of caution, you throttled the rollout. You published a policy instructing receivers to reject only ten percent of failing mail, expecting a gentle enforcement curve where only one in ten failing messages hits the wall. You watch the aggregate reports roll in over the next forty-eight hours and realize something is fundamentally broken across the internet's routing fabric. Your marketing platform is getting blocked completely at Google, passing untouched through Microsoft 365, and landing in a strange quarantine purgatory at a key vendor using Proofpoint.

This is the operational reality of dmarc pct tag behavior. The specification outlines a neat mathematical probability, but actual enforcement happens at the absolute discretion of the receiving Mail Transfer Agent. The protocol allows domain owners to request a specific handling mechanism, but it does not mandate how receivers implement the underlying randomization algorithms or how they layer those requests against their own proprietary threat detection engines.

Understanding the chasm between the RFC documentation and real-world receiver behavior is non-negotiable for security engineers managing domain reputation. If you rely on percentage tags to mask poorly aligned authentication infrastructure, you are entirely at the mercy of how different gateway vendors interpret your DNS records.

The Illusion of Mathematical Certainty in RFC 7489

The DMARC specification introduces the percentage tag as a mechanism to gradually phase in enforcement. The premise feels straightforward to analysts used to deterministic firewall rules. If you set a policy of p=reject and a pct of 25, the receiver is instructed to reject twenty-five percent of messages that fail DMARC evaluation. The remaining seventy-five percent fall back to the next least restrictive policy, which in this scenario is quarantine. If your baseline policy is quarantine, the fallback is none.

Domain Owners are advised to apply this mechanism with caution. The percentage is applied to the DMARC-failing message stream, not the total message stream. — RFC 7489 Section 6.1

That caveat trips up junior analysts constantly. The percentage math only initiates after SPF and DKIM have failed to produce an aligned pass. A common failure mode occurs when an administrator sees a ten percent reject rate in their aggregate reports and assumes only ten percent of their total traffic is being dropped, entirely missing that the calculation only applies to the subset of mail that already failed authentication.

The deeper architectural issue is that RFC 7489 leaves the exact mechanism of probabilistic enforcement entirely up to the implementer. The specification does not mandate a specific hashing algorithm, a time-based rolling window, or a globally synchronized counter. Receivers are left to figure out how to mathematically throttle failing emails at line speed across massive, globally distributed infrastructure. Because building synchronized state machines across planetary edge nodes is computationally wasteful, the major providers all chose completely different paths to achieve the same theoretical outcome.

Google Workspace and the Probabilistic Dice Roll

Edge Server Autonomy

Google handles incoming SMTP traffic at an incomprehensible scale. When an external server connects to a Google Workspace MX record, the specific edge node accepting that connection has to make a split-second routing and security decision. Maintaining a centralized database to track whether a specific sending domain has hit its exact ten percent threshold for the current hour introduces unacceptable latency. Instead of global tracking, Google relies on local probabilistic sampling.

Every time a message fails DMARC evaluation for a domain utilizing a percentage tag, the handling server effectively generates a random number. If the rollout is set to twenty percent, a generation of one through twenty triggers the requested reject policy. Twenty-one through one hundred triggers the fallback quarantine policy. Across millions of messages sent by high-volume transactional platforms, this local randomization smooths out statistically to match the requested percentage with high fidelity.

Low-Volume Variance

The architecture breaks down operationally when you are dealing with low-volume senders or highly targeted spear-phishing campaigns. If your shadow IT marketing tool only sends fifty emails a day to Gmail addresses, probabilistic sampling can result in wild daily variance. Monday might see zero rejections. Tuesday might see twenty. This variance makes troubleshooting incredibly difficult for SOC analysts relying on daily RUA reports. You cannot look at a single failed campaign and determine if the infrastructure is fixed, because the successful delivery might just be the result of a lucky probabilistic roll at the edge node rather than a valid DKIM body hash.

Microsoft Exchange Online Protection and the Quarantine Override

Microsoft Exchange Online Protection takes a decidedly different approach to handling partial DMARC enforcement. While M365 does parse and respect the core mathematical concept of the tag, its implementation is heavily subjugated by its own anti-spam architecture and tenant-level configurations. EOP does not treat DMARC as an absolute directive but rather as a highly weighted signal feeding into the broader Spam Confidence Level calculation.

When a message fails DMARC against a partial reject policy, M365 applies its own probabilistic logic to determine which bucket the message falls into. However, the exact behavior changes drastically depending on how the receiving tenant has configured their Anti-Phishing policies within Defender for Office 365. Microsoft allows tenant administrators to dictate exactly what happens to messages that fail DMARC, often overriding the sender-requested policy entirely. If an EOP administrator sets the organizational action for DMARC failure to move messages to the Junk Email folder, that local policy takes absolute precedence over your DNS records.

The Fallback Fallacy

The most critical divergence in Microsoft infrastructure involves the fallback mechanism. Under the specification, if a policy evaluates a failing message and lands in the un-enforced percentile, the message is supposed to be treated under the next least strict policy. If your intent was to precisely measure reject impact without affecting inbox placement for the fallback group, M365 routing will quickly disabuse you of that notion. EOP routing logic often treats these fallback quarantines identically to standard spam verdicts. The message is stamped with a failed Authentication-Results header and routed straight to the junk folder regardless of the mathematical outcome.

Secure Email Gateways and the Binary Reality of Proofpoint

Secure Email Gateways exist to enforce rigid, determinable security boundaries for enterprise networks. Probabilistic routing is inherently antithetical to how most SEG administrators build detection engineering pipelines. When a message hits a Proofpoint or Mimecast cluster, the organization paying for that appliance generally expects a binary outcome. It is either malicious or benign. It passes authentication or it fails. There is very little appetite for allowing unauthenticated mail through based on a coin flip.

Proofpoint administrators frequently configure their DMARC verification modules to completely ignore the percentage tag. The prevailing logic in highly regulated environments like finance and healthcare is that if a domain owner explicitly publishes a reject policy, any failure should be discarded to prevent BEC via lookalike domains or direct spoofing. The risk of allowing an unauthenticated payload through to a corporate inbox simply because the sender was trying to throttle their rollout is unacceptable to defensive architecture teams.

Authentication-Results: mx.example.com; dmarc=fail (p=reject dis=reject) header.from=sender.com

You will frequently see Authentication-Results headers exactly like that from SEG-protected infrastructure even when your policy clearly specifies a one percent rollout. The gateway evaluates the policy, sees the reject intent, drops the percentage parameter entirely, and terminates the SMTP connection. For security engineers analyzing aggregate reports, this creates a massive analytical discrepancy. Your reports from consumer mailbox providers show a gentle curve of enforcement, while your B2B traffic to enterprise domains shows absolute devastation.

Architecting a Resilient Enforcement Strategy

The operational reality is that the percentage tag is a blunt instrument masquerading as a scalpel. Treating it as a precise control mechanism for a complex identity architecture will invariably lead to dropped legitimate mail and frustrated business units. Defensive architecture requires building your rollout strategy around the worst-case interpretation of the specification, assuming that any policy containing a reject directive will be interpreted as a total reject by enterprise gateways.

Eliminating Probabilistic Reliance

You cannot rely on percentage throttling to save poorly configured infrastructure. Before moving off a monitor policy, every known sending source must be definitively aligned. You must verify that SPF includes are not breaking under the ten DNS lookup limit and that DKIM body hashes match the payloads after edge servers append mandatory disclaimers. If you have legacy forwarding architecture that routinely breaks SPF alignment, you need to implement Authenticated Received Chain protocols defined in RFC 8617 before you even consider touching the enforcement policy.

The Subdomain Isolation Tactic

Instead of relying on mathematical tags to throttle a reject policy, throttle the rollout via subdomain isolation. Apply strict enforcement policies to your primary corporate domain and explicitly block sending from unregistered subdomains. Migrate your legacy applications, third-party marketing vendors, and automated notification systems to dedicated subdomains. You can then manage their DMARC policies independently. This isolates the blast radius of an authentication failure to a single business unit or vendor, rather than risking your primary corporate mail flow on a probabilistic dice roll at a remote edge server.

Closing the Visibility Gap

The specification gives senders the illusion of control, but the receiving MTA ultimately owns the routing table and makes the final determination. Understanding how Google Workspace applies probabilistic sampling, how M365 filters through tenant overrides, and why enterprise gateways force binary outcomes is the difference between a quiet security upgrade and an accidental denial of service against your own organization. You cannot math your way out of bad email authentication.

Stop treating aggregate reports as theoretical exercises and start reading them as operational telemetry. When you are ready to stop guessing how receivers are interpreting your DNS records, MailSleuth.AI can ingest those raw RUA streams and pinpoint exactly which gateways are ignoring your rollout strategy. The truth is always written in the headers, provided you have the tooling to read how the receiver actually wrote them.

The takeaway

The specification gives senders the illusion of control, but the receiving MTA ultimately owns the routing table and makes the final determination. Understanding how Google Workspace applies probabilistic sampling, how M365 filters through tenant overrides, and why enterprise gateways force binary outcomes is the difference between a quiet security upgrade and an accidental denial of service against your own organization. You cannot math your way out of bad email authentication.

Stop treating aggregate reports as theoretical exercises and start reading them as operational telemetry. When you are ready to stop guessing how receivers are interpreting your DNS records, MailSleuth.AI can ingest those raw RUA streams and pinpoint exactly which gateways are ignoring your rollout strategy. The truth is always written in the headers, provided you have the tooling to read how the receiver actually wrote them.

#dmarc#email-forensics#identity#detection-engineering#rfc-7489
MailSleuth Research
Email Security Team

We dissect phishing campaigns and email infrastructure so you don't have to.