Blazalek.com

ESP migration checklist: what to record and when to stop

Before shifting an email stream to a new ESP, you need clear criteria for whether to proceed, hold, or trigger a rollback review. This worksheet logs DNS authentication, suppression preservation, pending in-flight queues, and recovery gates, backed by a worked example of a cutover replay.

By

On this page

This ESP migration checklist resolves one question for one sending stream: can it proceed to the new ESP, must it stop, or does it need rollback review? Complete your plan before moving production traffic. Authentication, suppression continuity, and outstanding sends are separate questions, each needing its own evidence.

Provider documentation was checked on 14 September 2026; recheck provider capabilities before your own migration, since limits and features change. For root-cause diagnosis and traffic pacing, the full ESP migration procedure covers what this sheet records.

Setting up the record

Copy the blank templates below and add rows as needed. An empty cell means nobody has filled it in yet, which is different from not knowing. Write UNKNOWN for missing evidence, along with its owner and the next decision checkpoint. Use N/A only with an explicit operational reason. Store evidence references and timestamps here; credentials and recipient lists are stored elsewhere.

The context table identifies the stream, its business purpose, both ESPs with their stream IDs, and the planned window with its timezone. Record the visible From domain and, for each route, the SPF identity and DKIM d=/s=. Name the people: business decision-maker, integration or queue owner, DNS owner, and the owner of suppressions and events. Reference the baseline scope and window, and the preserved old and new configuration versions.

Stream / purposeOld → new ESP / stream IDsFrom; old → new SPF and DKIM identitiesWindow / timezoneNamed owners and rolesBaseline / configuration evidence

For each stage, name a specific checkpoint and fill the remaining columns:

  • Criterion: states what must hold, for which recipients or providers and over which observation window, before you collect results.
  • Observation: records the result and a dated evidence reference.
  • Owner: names the person who gathers evidence or resolves the issue.
  • Decision: records GO, HOLD/STOP, or ROLLBACK-REVIEW, with the reason and decision-maker.
  • Recovery: names the corrective action or escalation contact, and the next check time.

Rollback review does not authorize replay.

Stage / checkpointCriterion / scope / windowObservation / dated evidenceOwnerDecision / reason / decision-makerRecovery / escalation / next check
Preflight
Cutover
Rollback
Verification

Gathering evidence in dependency order

Preflight: routes and authentication

Before modifying outbound routing, preserve configuration versions, provider limits, and access to the old route; once traffic moves, you may need all three to compare headers or roll back. For Postmark, record the account’s domain-specific DNS values and verification status. Make sure the domain publishes exactly one selected SPF record and that the DNS-querying-term limit across evaluation holds, nested evaluation included. Publish distinct DKIM selectors for the new platform, and document how long the old public keys must remain published for mail still in transit.

For each route, preserve received samples showing visible From, checked SPF domain and result, and DKIM d=/s= and result, following the authentication evidence fields. Establish the receiving system’s Authentication-Results trust boundary; a matching authserv-id string alone is insufficient. Check alignment in the mode your DMARC policy sets. DMARC requires at least one aligned authenticated identifier, and passing authentication does not guarantee inbox placement.

Preflight: will the suppression list survive the move?

Record export coverage across unsubscribes, permanent hard bounces, spam complaints, and manual suppressions by stream. Document destination mappings, import results, and unresolved entries. Postmark’s suppression interface is stream-scoped; creating an entry by address only does not establish that the historical reason was preserved, and an HTTP 200 can still contain individual failures. Keep reason, origin, time, and stream history separately.

Assign someone to reconcile changes made since the copy, including late events from either route and any changes made before a failback. Postmark’s subscription events distinguish deactivation from reactivation, and ManualSuppression can mean an unsubscribe, so do not infer permission to reactivate from that label. The gate I recommend: mark the affected recipient set HOLD/STOP until missing coverage, mappings, or failures are resolved. The guide covers provider suppression versus your own suppression list.

Cutover

The business and integration owners choose stage size, observation window, and tolerable outcomes for this stream. Record the activated configuration version and activation time. Compare received identities and recipient-provider outcomes with the scoped baseline, stopping ramp-up if a required criterion fails or evidence is missing. These are worksheet controls, not universal rollout thresholds; Postmark’s event meanings help interpret what you observe.

Track each outstanding logical send and recipient separately. Enter the provider or message ID, the state with timed evidence, the person controlling retries and queues, and the chosen disposition with decision-maker and time. Distinguish NOT SUBMITTED, ACCEPTED/QUEUED, CONFIRMED OUTCOME (name which outcome), and UNKNOWN. In Postmark, Queued means accepted but not yet sent, Processed means Postmark has handed it to its mail servers, and Delivered means the recipient server accepted the message, which is not inbox placement. These states are not interchangeable.

Logical send / recipient referenceProvider / message IDState / timed evidenceRetry / queue ownerDisposition / decision-maker / time

Rollback: which sends can be replayed?

Pause expansion and reconcile outstanding attempts first. A positive final SMTP reply transfers delivery responsibility to the receiving server; a final-reply timeout can leave acceptance uncertain. Exclude accepted or unknown sends from blanket replay, and escalate unresolved decisions about duplicate-versus-delay risk. Postmark documents no idempotency-key support, and application safeguards cannot guarantee deduplication across ESPs. The guide chapter on sending reliability describes how an idempotency key works.

Before you route future eligible sends back, record the version to restore, the person authorized to restore it, the old route’s authentication and capacity readiness, and current suppressions. Restoring a version does not prove the queues were cancelled, and it does not prove that mail can be recalled after recipient-server handoff. Obtain provider evidence for cancellation or expiry; otherwise the item stays UNKNOWN. Retain the old DKIM public keys for the overlap period you have justified.

Verification

Collect dated evidence for received samples, recipient-server outcomes, application and event processing, and suppression enforcement. Handle repeated events without repeated side effects; webhook retries and pauses need a reconciliation owner, and a webhook retry is not an email retry. The guide explains why a webhook handler has to be idempotent.

When you close the record, name the evidence approver, accepted exceptions, remaining owners, and criteria for retiring the old route. The planned cutover time alone does not retire dependencies.

A synthetic example: authentication passes, replay stops

The fictional inputs below describe EX-M1, a receipt stream moving from ESP A over SMTP to Postmark, with application revision r18 → r19. "Receipt owner" makes business decisions; "Integration owner" controls queues. Their local gate requires a known disposition for every affected earlier attempt before replay, and they choose T+30 minutes for the review.

E-AUTH: At T+2, a supplied new-route sample has From: billing@receipts.example.com, a DKIM pass with d=receipts.example.com; s=pm2026, an SPF pass for MAIL FROM domain bounce.example.net, and a DMARC pass. Its applicable policy has adkim=s; aspf=s. The scenario explicitly establishes Authentication-Results as originating from the trusted recipient system.

E-QUEUE: At T−1, the application attempted to send receipt-1042 to customer@example.org through ESP A. The client timed out after DATA while waiting for the final reply. No positive completion reply, final event, or provider confirmation is available, and this logical send has not been submitted to Postmark.

Stage / checkpointCriterion / scope / windowObservation / dated evidenceOwnerDecision / reason / decision-makerRecovery / escalation / next check
Cutover / receipt-1042 replayKnown disposition before replay; review T+30E-AUTH at T+2 passes authentication; E-QUEUE at T−1 remains UNKNOWNIntegration ownerSTOP replay; hold item and expansion under Receipt owner’s gateObtain ESP A outcome evidence; Receipt owner reviews duplicate/delay impact at T+30

To reconstruct the row, compare the DKIM signing domain and the SPF MAIL FROM domain with receipts.example.com. Only DKIM meets strict domain equality. That is enough for DMARC, but it proves only this one sample’s authentication condition.

E-QUEUE cannot establish non-acceptance: a final SMTP reply timeout can precede a duplicate on retry. The stated gate therefore stops the replay. T+30 is a chosen decision time, not a queue expiry and not a safe retry time. If the evidence is still missing then, retain UNKNOWN and escalate.

Related procedures: the sender inventory, Postmark domain setup, and the DNS change worksheet.

Frequently asked questions

Does a Postmark Delivered event mean the message reached the inbox?

No. In Postmark, Delivered means the recipient server accepted the message, not that it reached the inbox. Queued means accepted but not yet sent, Processed means handed to its mail servers, and these states are not interchangeable.

Does a rollback review allow replaying sends?

No. Rollback review does not authorize replay. Exclude accepted or unknown sends from any blanket replay, and escalate unresolved decisions about duplicate-versus-delay risk.

Can you replay a send that timed out waiting for the final SMTP reply?

Not on that evidence alone. A final-reply timeout can leave acceptance uncertain, and a final SMTP reply timeout can precede a duplicate on retry. In the example, the stated gate therefore stops the replay.

Does copying suppressions by address preserve the original reason?

Creating an entry by address only does not establish that the historical reason was preserved, and an HTTP 200 can still contain individual failures. Keep the reason, origin, time and stream history separately.

Is a webhook retry the same as an email retry?

No. A webhook retry redelivers an event notification to your application, not an email to a recipient. Webhook consumers must handle duplicate event payloads idempotently.

Email deliverability consultant

Get a second opinion on your email setup

Tell me what your customers are seeing or what you’re planning to change. In 30 minutes, we’ll work out what your team can check first and whether a deliverability audit would help.

I led deliverability at GetResponse, designed a dedicated newsletter system for Easy.Tools and helped Youmoney migrate its automated emails in stages. Read the case studies

Book free call

Email deliverability audit

See what I check and read a sample report with recommendations for your team.