Octeth v6.0.7 Now Available! See What's New

Stage 10 of 12

Deliverability Operations for an Email Service Provider

Once customers send, deliverability is daily work: reading what Gmail, Microsoft and Yahoo report, getting complaints back into Octeth, throttling per provider, classifying bounces and triaging blocklists. This guide covers how we run it on PowerMTA and what the provider tools show today.

By Cem Hurturk, founder of Octeth Updated 17 min read

Skills
  • Bounce and complaint processing
  • Postmaster tools
  • Blocklist checks
On this page
  1. Watch domain reputation first, IP reputation second
  2. Google Postmaster Tools
  3. Gmail Feedback-ID: campaign data, not a per-message loop
  4. Microsoft SNDS and JMRP
  5. Yahoo Complaint Feedback Loop: keyed on the DKIM domain
  6. Get complaints into Octeth
  7. Throttle and back off per provider
  8. Classify bounces
  9. Blocklist triage: Spamhaus first, then the filters your recipients use
  10. Seed tests and their limits
  11. Opens after Apple Mail Privacy Protection
  12. A weekly routine

Setup ends when the first customer sends. From then on, deliverability is operations: reading what mailbox providers report, getting complaints back into Octeth, slowing down when a provider asks, and finding the one customer behind a block. The sending setup is in Stage 5, sending infrastructure and IP warm-up, authentication and the bulk sender rules are in Stage 6, sender domain authentication, and the complaint and bounce limits that suspend an account are in Stage 8, abuse prevention. This stage is what you watch and do every week.

No setup in this guide guarantees inbox placement. It gives you the data to find problems early and the controls to contain them.

Watch domain reputation first, IP reputation second

Mailbox providers judge the domains in a message as well as the IP it comes from. Gmail’s rate-limit replies name the IP, the IP netblock, the DKIM domain, the SPF domain or a URL domain in the message as the source of the problem (Gmail SMTP errors). Its blocks name either the sending IP or the sending domain as having very low reputation.

On one shared MTA we measured, the largest single Gmail block reason in a day was the low reputation of the sending domain, ahead of the IP rate limit. That matches what an ESP sees: each customer’s mail is signed with the customer’s domain (Stage 6, sender domain authentication), so a customer carries its domain’s reputation onto any IP you give it. Moving that customer to a fresh IP leaves the domain problem in place and spends the IP.

So read problems in this order:

  1. Which customer and domain. Group bounces by account first. On shared IPs one customer is usually the cause.
  2. Which IP. If every customer on an IP is blocked at one provider, the IP or its range is the problem.
  3. Which provider. A block at one provider with clean delivery elsewhere points at that provider’s data, not a public blocklist.

Authentication is required but does not fix reputation. On the same MTA, one sender passed SPF and DKIM and still bounced about 95% of its mail, because Microsoft had banned its IP and its list looked bought.

Google Postmaster Tools

Postmaster Tools shows Gmail’s view of mail to personal Gmail accounts, per verified domain. Verify your infrastructure domain, and ask customers who send in volume to verify their own domains and share access.

Mailbox provider feedback sources Gmail: Postmaster Tools is keyed on a verified domain and shows daily rates such as spam rate, authentication and delivery errors in its dashboard. Gmail's Feedback-ID loop is keyed on the identifiers in the Feedback-ID header and shows spam rates per identifier in the Postmaster Tools dashboard, never which address complained. Microsoft: SNDS is keyed on the sending IP and shows a daily filter result and complaint rate per IP in its portal. JMRP is keyed on the sending IP and sends a report for each junk complaint. Yahoo: the Complaint Feedback Loop is keyed on the DKIM d= domain and sends an ARF report for each spam complaint. Only the JMRP and Yahoo reports name the message, and they reach Octeth's suppression list through an email pipe, a POP3 mailbox, or the MTA's accounting log posted to Octeth's log API command, not the bounce webhook. Gmail complaints do not suppress anyone in Octeth by themselves. Gmail Postmaster Tools Dashboard Keyed on a verified domain Daily rates: spam, authentication, errors Feedback-ID Dashboard Keyed on header identifiers Spam rates per identifier, no addresses Microsoft SNDS Dashboard Keyed on the sending IP Daily filter result and complaint rate JMRP To Octeth Keyed on the sending IP A report for each junk complaint Yahoo Complaint Feedback Loop To Octeth Keyed on the DKIM d= domain An ARF report for each spam complaint Octeth suppression list Email pipe, POP3 mailbox or the log API
Only Microsoft JMRP and Yahoo's loop report each complaint, so only they can suppress an address in Octeth. Gmail reports rates, never which address complained.

Version 2 status, checked 19 November 2026

  • The new interface (v2) launched in 2024. Google asks all senders to move to it. It has a Compliance status dashboard that checks the sender guidelines.
  • The old interface is not gone yet. Google has postponed the deprecation of the legacy web interface and says it will share timing later.
  • Domain and IP reputation are being retired. All old dashboards move to v2 except Domain Reputation and IP Reputation. Google says reputation data was hard to act on, slow to reflect changes and misleading on its own, and plans new dashboards in their place.
  • The v1 API is being retired. The v2 API is available. It covers v1’s data except domain and IP reputation and adds compliance status. It uses a different schema, so integrations need code changes.

Source: Learn about the deprecation of the old Postmaster Tools interface. Read Postmaster Tools in Google’s web interface. Check that any integration you rely on calls the v2 API.

What to read

DashboardWhat it tells an ESP
Compliance statusWhether the domain meets the sender guidelines: SPF and DKIM, DNS, message format, TLS, spam rate, and for bulk senders DMARC, one-click unsubscribe and honouring unsubscribes. Changes can take up to 7 days to show.
Spam rateThe share of mail delivered to the inbox that users then marked as spam. Keep it below 0.10% and never reach 0.30%.
Feedback loopSpam rates per Feedback-ID identifier. See the next section.
AuthenticationSPF, DKIM and DMARC pass rates.
Delivery errorsRejections and temporary failures as a share of authenticated mail.

Sources: Postmaster Tools dashboards, Email sender guidelines.

A low spam rate is not always good news. Google says the rate counts only complaints on mail that reached the inbox. If Gmail already sends much of a domain’s mail to spam, few recipients see it to complain, and the rate looks low or drops to zero. Read the spam rate next to delivery errors and your own bounce text.

Data is daily, in UTC, usually updated within 24 hours, and missing on days with low volume.

Gmail Feedback-ID: campaign data, not a per-message loop

Gmail does not send you a report per complaint. Its feedback loop reports spam rates for identifiers you put in a Feedback-ID header, and only when an identifier has an unusual spam rate across enough mail and enough distinct reports. The data covers @gmail.com recipients only and shows in the Postmaster Tools Feedback loop dashboard (Feedback Loop).

So Gmail tells you which customer or campaign draws complaints, never which address complained. Gmail complaints do not suppress anyone in Octeth by themselves. Suppression comes from list unsubscribes and the provider loops below.

The header has up to three identifiers of your choice and a mandatory sender ID of 5 to 15 characters at the end:

Feedback-ID: <campaign>:<customer>:<mail type>:<sender id>

Four rules decide whether it works for an ESP:

  • Sign with a domain you control. The header must be covered by a DKIM signature from a domain that the sender owns and has verified in Postmaster Tools. Mail signed only with each customer’s domain puts the data under each customer’s domain, not yours.
  • Make the customer field unique. Google aggregates each field on its own, so a campaign ID that two customers share mixes their data. Never use a value unique to each message.
  • One header per message. Do not let a customer’s own Feedback-ID pass through next to yours.
  • SPF and PTR for the sending IPs. The IPs must be in the SPF record of the signing domain, and each must have a PTR record that resolves to a valid hostname, preferably one of the DKIM domains.

Stage 5 adds the X-Job header on the staff Email delivery screen. A Feedback-ID can be added there, or in a user group’s custom headers, before you open signups. Two things to test first: the gateway API passes a customer’s own headers through, and a user group’s Remove list also strips a header the same group adds. Send a gateway message that carries its own Feedback-ID and check what arrives.

Microsoft SNDS and JMRP

Microsoft’s programs work per IP, not per domain. Both live in the Smart Network Data Services portal (SNDS).

  • SNDS shows, per IP and per day, mail volume, the share of messages its filter judged spam (green under 10%, yellow between, red over 90%), the complaint rate and whether the IP is blocked or its mail is junked. IPs that sent fewer than 100 messages that day may show no data. Data is kept for 90 days, and an automated download URL returns the same CSV as the export button (SNDS FAQ).
  • JMRP sends you a report with selected message headers when an Outlook.com user marks your mail as junk. Microsoft says feedback can start within as little as 72 hours of enrolment (Services for senders). On a shared IP you see every complaint for that IP, whichever customer sent it.

SNDS announcements, last updated 22 September 2026, describe a new portal and say that access no longer needs periodic reattestation. The FAQ still describes a 10-month reattestation. The announcement is newer.

Getting access to leased IPs

SNDS grants access by email. For an IP range, it sends the authorization link to postmaster@ or abuse@ of the domain in the range’s PTR names, or to the contacts in the range’s RDAP record. The PTR method works for ranges up to a /23. Make sure postmaster@ and abuse@ on your PTR domain receive mail. If your IP provider set the PTR names, or only the provider is in RDAP, the link goes to the provider, so arrange who forwards it before you need the data.

Pull SNDS every day through the automated URL and store each IP’s filter colour with its history, so a yellow IP is seen before it turns red.

When Microsoft blocks an IP

A reply like 550 5.7.606 Access denied, banned sending IP means the IP is on Microsoft 365’s blocked senders list. Microsoft 365 has a delist portal at sender.office.com, and Outlook.com consumer mail has a separate delisting form (Microsoft delist portal). Stop the sender first. Stage 5, sending infrastructure and IP warm-up, records a banned sender that was moved to a fresh IP and got that one banned too.

Yahoo Complaint Feedback Loop: keyed on the DKIM domain

Yahoo’s loop is domain-based and works only for DKIM-signed mail. You create a sender profile, verify a domain and enrol it. When a message signed with an enrolled DKIM key is marked as spam, Yahoo sends an ARF report to your address (Yahoo Complaint Feedback Loop).

For an ESP this is the opposite of Microsoft. JMRP follows your IPs. Yahoo’s loop follows the d= domain in the signature. If your MTA signs each customer’s mail with the customer’s domain, as in Stage 6, sender domain authentication, Yahoo reports go to whoever enrolled that customer’s domain. Decide before launch how you will get them:

  • Enrol customer domains yourself as part of sender domain setup, with each customer’s agreement.
  • Add a second DKIM signature with a domain you own, so one enrolment covers every customer. Test that your MTA supports it before you rely on it. PowerMTA 4.0 signs once per message. Its dkim-identity setting swaps the signing domain instead of adding a second signature, so on that version the choice is between the customer’s domain and yours.

The same choice decides whose Postmaster Tools account sees the Feedback-ID data. Make it once for both.

Get complaints into Octeth

A complaint helps only when it reaches Octeth and suppresses the address. Octeth reads complaints from an email pipe or a POP3 mailbox, where it identifies the subscriber from the X-MessageID header in the report (Octeth complaint processing). PowerMTA logs complaints as f records in its accounting log. PowerMTA writes an f record only for feedback loop mail its feedback loop processor receives. Those records go in through the process_pmta_log_file admin API command, not through the bounce webhook: in 6.0.7 the webhook ignores f rows on the gateway path and would register them as hard bounces on the campaign path. A log shipper set up for bounces forwards no f rows, so set up the complaint path on its own and test it. Octeth’s complaint page says the webhook accepts f rows. This guide follows the 6.0.7 code. Stage 8, abuse prevention, covers what the suppression does.

Then prove it. Mark a test message as spam at Outlook.com and at Yahoo, and check that the address lands on the suppression list. Repeat the test complaint after every MTA, log shipper or Octeth change. Complaints are rare, so an old newest-complaint time in Octeth is normal and proves nothing. For email gateway sends (from Octeth 6.0.1, with the gateway feedback monitor on), Octeth’s Bounce Processing page has an MTA feedback box that shows the age of the newest delivered and bounced events. An age there that keeps growing while customers send means the feedback path is broken.

Throttle and back off per provider

Each provider tells you to slow down in its own words. The MTA must hear it and act per provider, not globally. A PowerMTA setup that does this:

  • Connections are capped per provider. Set max-smtp-out in a <domain> block for each large provider, such as Gmail, Outlook, iCloud and Yahoo. PowerMTA’s default is 20 per domain. Lower it for a provider that replies with too many connections, and raise it only while that provider’s deferrals stay flat.
  • Messages are rate-limited per domain. Set a default rate per domain, and stricter rates for providers that throttle hard.
  • Replies map to actions. A pattern list puts a queue into backoff on rate-limit replies, bounces the whole queue on a reputation block, and bounces a single recipient at once on a temporary reply to that recipient, such as a full mailbox.
  • Backoff is slow on purpose. backoff-retry-after and backoff-max-msg-rate set how a queue in backoff retries. PowerMTA’s defaults are one hour and no rate cap, so set a low backoff rate per provider, or a throttled queue resends at full speed each time it retries.

A pattern list of that shape, with reply text from Gmail’s error list and a generic retry message:

<smtp-pattern-list provider-replies>
    reply /4\.7\.28/                   mode=backoff
    reply /try again later/            mode=backoff
    reply /very low reputation/        bounce-queue
</smtp-pattern-list>

<domain *>
    smtp-pattern-list provider-replies
</domain>

PowerMTA uses the first matching rule. bounce-queue bounces every message waiting in that queue, and bounce-rcpt works only on 4xx replies to RCPT. Patterns are case-insensitive regular expressions. The PowerMTA 4.0 user guide, section 3.3.17, lists every action.

Two decisions matter more than the exact values:

  • What a bounced queue does in Octeth. A queue bounced on a reputation block produces bounces for good addresses. Make sure those do not suppress them. The next section covers it.
  • How long deferred mail waits. Retry interval and expiry decide whether a throttle ends in delivery or in a wave of expired mail. Stage 5, sending infrastructure and IP warm-up, records a Microsoft throttle where most deferred mail expired before the throttle lifted.

Classify bounces

Bounce rate alone misleads. On one shared MTA, on a day when more than four in ten messages bounced, most bounces came from providers refusing the mail. Bad and inactive addresses were a small share. Cleaning lists would not have touched the problem.

The reply code gives the first split. In an enhanced status code, class 4 is a temporary failure and class 5 a permanent one. The middle digit says what failed: 1 is the address, 2 the mailbox, 7 security or policy (RFC 3463). PowerMTA adds a category to each bounce in its accounting log.

What failedExampleAction
The address5.1.1 user unknown, category bad-mailbox or bad-domainSuppress the address. A spike from one account means a bad import.
The mailbox4.2.2 mailbox full, category quota-issuesRetry. Octeth turns repeated soft bounces into a hard bounce, 3 by default.
Policy or reputation5.7.1 low reputation, 5.7.606 banned IP, category spam-relatedDo not judge the address. Find the account, IP and provider, and read the reply text.
Authentication5.7.26 unauthenticatedFix the customer’s records. See Stage 6, sender domain authentication.
Throttling421 4.7.28 rate limited, then message-expiredBack off. Expired mail means the queue never cleared.

Octeth classifies each bounce as hard or soft by matching the reply text against more than 600 built-in patterns, and you can add your own (Octeth bounce processing). On the campaign webhook path Octeth 6.0.7 treats every row it receives as a hard bounce, whatever the category. Your log shipper must send only bad-mailbox, bad-domain, invalid-mailbox and inactive-mailbox, or a block on a shared IP suppresses thousands of good addresses. Gateway rows are filtered to those four inside Octeth.

Only bounces that reach Octeth count. With customer return paths, an MTA that accepts inbound mail only for its own domains sees only bounces returned during delivery (Stage 5, sending infrastructure and IP warm-up).

Blocklist triage: Spamhaus first, then the filters your recipients use

There are many blocklists. Rank them by how much of your recipients’ mail they decide.

Spamhaus

Spamhaus has several lists, and each has its own removal rule:

ListListsRemoval
SBLIPsRemoval requests must come from the ISP in charge of the listed IP, after the problem is solved, and Spamhaus asks how it was solved. For leased IPs that is your IP provider’s abuse desk (SBL).
CSSIPsListings usually expire three days after the last detection. Self-removal is allowed within limits, and the IP is listed again at once if the problem continues (CSS FAQ).
DBLDomains, in the sender or in linksRequest removal from the lookup page. A domain listing follows the sender to any IP (DBL FAQ).

Look up listings at check.spamhaus.org, or by DNS from a resolver Spamhaus answers. Spamhaus refuses queries from public and open resolvers and answers them with 127.255.255.254, which is an error, not a listing or a clean result (DNSBL usage FAQ). Spamhaus offers a free Data Query Service for senders who cannot use their own resolver. Never request removal before the sender is fixed, or the listing returns.

Corporate filters and minor lists

Business recipients sit behind filtering services with their own reputation data. On one shared MTA, Proofpoint rejected nearly all mail to Proofpoint-hosted recipients from every main IP over the days we measured, and kept blocking those IPs for weeks, while consumer providers delivered part of the same mail. Their data is not in the public DNS blocklists. Use their lookup pages:

  • Proofpoint: ipcheck.proofpoint.com
  • Cloudmark: csi.cloudmark.com
  • invaluement: invaluement.com/lookup
  • Microsoft 365: sender.office.com

invaluement’s ivmSIP/24 lists whole ranges, as Stage 5, sending infrastructure and IP warm-up, describes. Rotating senders inside a listed range changes nothing.

Lists that no provider you mail uses cost you nothing, and a listing on one does not explain a block. On the same MTA, Barracuda listed IPs that delivered well and did not list the IP Microsoft had banned. Start from the bounce text, which names the blocker, and look up only the lists your recipients’ servers actually used.

Triage order

  1. Count bounces per blocker, per IP and per account from the MTA log.
  2. Find the account behind the blocked mail. On shared IPs, fix or stop that account first.
  3. Check the IP or domain on the list named in the bounce text.
  4. Request removal after the cause is gone, through the channel that list requires.
  5. Write down what was listed, when, why and when it cleared.

The MX record inspector shows which provider or filter hosts a recipient domain, so you know whose rules apply.

Seed tests and their limits

A seed test sends a campaign to mailboxes you control and reports where each one landed. Use it to catch setup errors before a launch: a broken signature, a missing unsubscribe header, a link domain that filters dislike. Stage 12, Operations and launch, covers the pre-launch run.

A seed shows where mail landed in that one seed mailbox at that moment. It says nothing about the customer’s real recipients and how they engaged with past mail. Treat a good seed result as “nothing is broken”, not as inbox placement for the list.

Seeds also add noise. On a shared MTA where the sending software added seed addresses to every send, the seed addresses had no subscriber record, so the bounce endpoint rejected their bounces by design and logged them as errors. One seed domain deferred thousands of messages a week. Filter seed traffic out of your reports and alerts.

Opens after Apple Mail Privacy Protection

With Protect Mail Activity on, Apple Mail hides the reader’s IP and downloads remote content in the background when a message is received, not when it is viewed. Senders then cannot see when or how many times a message was opened (Apple Mail privacy settings). The tracking pixel loads for mail the person may never read, so opens from Apple Mail users are inflated.

Gmail says it does not track open rates, cannot verify open rates reported by third parties, and that low open rates are not necessarily a sign of deliverability problems (Gmail sender guidelines).

For an ESP that means three things:

  • Do not use opens as a health signal. Watch bounces, deferrals, blocks and complaints per account and per IP instead.
  • Do not build engagement segments on opens alone. Use clicks and other actions, or warm-up and re-engagement segments will keep contacts who never read.
  • Explain it to customers. Opens can mislead both ways. The low-open account in the pitfalls below had an IP ban, not bored readers.

A weekly routine

Run a read-only scan of the last seven days of the MTA’s accounting log several times a week. Each run looks at:

  • outcomes per account and per IP: attempts, delivered and bounced
  • bounces per blocker, using the reply text, with an unmatched bucket that someone reads
  • new accounts and new sending domains
  • deferrals and expired mail per provider
  • the bounce canary and the age of the newest delivered and bounced events in Octeth

Add the provider tools to it: the SNDS colours for every IP, the Postmaster Tools compliance status and spam rate for every domain you can see, and complaint counts per account. Each run ends with a dated note of what changed and what is still open.

Before you move on

  • Your infrastructure domain and every sending IP are set up in the provider tools you can use: Postmaster Tools, SNDS and JMRP, and Yahoo's Complaint Feedback Loop.
  • You know which domain signs each customer's mail, and so whose account in each tool holds that customer's data.
  • A test complaint from each feedback loop reaches Octeth and suppresses the address.
  • Your MTA backs off on provider rate-limit replies, and you know how long deferred mail may wait.
  • You know which bounce categories suppress an address, and a reputation block does not.
  • You check blocklists from a resolver Spamhaus answers, and you know the lookup page for each corporate filter your recipients sit behind.
  • A report shows bounces per account, per IP and per blocker for the last seven days, and someone reads it several times a week.
  • No dashboard, report or customer promise relies on open rates alone.

What we've seen go wrong

A low open rate that was an IP ban

A customer on a shared MTA we run asked why one account opened at 1 to 5% while its sibling accounts opened at 25 to 50%. The cause was delivery, not engagement. Microsoft had banned that account's IP, and most of its mail that month was never delivered. The banned IP was clean on Barracuda, SpamCop and SORBS, while sibling IPs that delivered better were listed on Barracuda. Read the bounce text per IP before you read opens or public blocklists.

A complaint feed that reported nothing

On a shared MTA, a full day of heavy sending logged zero complaint records. With that volume and a high bounce rate, zero complaints was not good news. Either the feedback loop addresses were not receiving reports, or providers were blocking instead of delivering. Treat a silent feedback loop as a fault until a test complaint proves otherwise.

A domain listing that followed the sender

Spamhaus DBL listed one sender's domain on a shared MTA. The bounce text named the domain in the links and in the sender address. A domain listing is not tied to an IP, so moving that account to another IP would not have helped. Most of those bounces came on one day. By the time we looked, a live lookup could not confirm the listing. Keep the bounce text with its timestamps as the record.

Blocks that our report did not count

Our weekly scan of a shared MTA's log groups bounces by known blocker text. It had no pattern for Microsoft's banned-IP reply or for invaluement's ivmSIP, so a week of those bans and rejections was missing from its blocker table. They showed up only when we drilled into the raw rows. Add a pattern the first time you see a new blocker, and check what lands in the unmatched bucket.

A blocklist answer that meant nothing

Checking a Spamhaus DBL listing from a server that used a public resolver returned 127.255.255.254. Spamhaus returns that code for queries through public or open resolvers, so it is no answer, not a clean result. Use a resolver Spamhaus answers, or its free Data Query Service, or the web lookup at check.spamhaus.org.

We build ESPs

Rather have us build this?

Deliverability consulting is part of the managed ESP: we advise on reputation, authentication, blocklists and mailbox provider rules.