Handling bounces and complaints: DSNs, enhanced status codes and feedback loops

Goal

Stop sending to addresses that cannot or do not want to receive mail, automatically and quickly, so that sender reputation and recipients' trust are preserved.

Prerequisites

A machine-read return path (a dedicated bounce mailbox or the provider's bounce events), a suppression table keyed by address and stream, and the two report formats: delivery status notifications (RFC 3464, a message/delivery-status part inside multipart/report) and abuse feedback reports (RFC 5965, a message/feedback-report part).

Steps

1. Route bounces to the machine: set the envelope sender (Return-Path) to a tagged address per message or stream (bounce+order-1234@…) so a bounce can be matched even when the DSN omits the original headers.
2. Parse the DSN's per-recipient fields: Final-Recipient, Action (failed, delayed, delivered, relayed, expanded), Status and Diagnostic-Code. Status carries an enhanced status code class.subject.detail.
3. Classify by class. RFC 3463 defines 5.X.X as a permanent failure, not likely to be resolved by resending, and 4.X.X as a persistent transient failure where sending later may succeed. The subject subcode says where: X.1.X addressing (5.1.1 bad destination mailbox), X.2.X mailbox (4.2.2 mailbox full), X.7.X security or policy (5.7.1 delivery not authorized).
4. Suppress immediately on 5.1.X and other permanent address errors. Count 4.X.X per address and suppress after a policy threshold. Treat X.7.X policy rejections as a signal about the sender's reputation or authentication, not as evidence about the address.
5. Register with mailbox providers' feedback loops and parse ARF reports: the second MIME part (message/feedback-report) carries Feedback-Type (abuse, fraud, virus, other), the third part the original message or its headers. Suppress on the first abuse report; never reply to the complainer.
6. Store every event with the parsed code, the raw diagnostic text and the message reference; alert on bounce and complaint rates per stream, not only in total.
7. Keep non-standard replies (out-of-office, free-text rejections) out of the suppression logic unless the provider classifies them; a heuristic that suppresses on "undeliverable" in the subject line will misfire.

Expected result

A suppression list that grows from evidence, low bounce and complaint rates, and an answer to "why did this recipient stop getting mail?" within one query.

Limits and test basis

Not every receiver sends standard DSNs or offers a feedback loop; some large providers expose only aggregate spam rates. Soft-bounce thresholds are policy, not standard. Format facts follow the cited RFCs; no rates are claimed.