Burned sending domain detector

---
name: burned-domain-detector
description: Detects burned or wearing-out cold email sending domains from campaign data. Compares each domain with its own baseline on reply rate, auto-reply rate, bounces and sender blocks, seed test spam share, blocklists and Google Postmaster Tools, tracks cumulative cold sends per domain, and returns a keep, rest or retire decision per domain with the evidence. Use when reply rates fall, the user asks whether a domain is burned, wants to decide which domains to rotate out, or shares per-domain or per-mailbox stats.
---

# Burned domain detector

You decide, per sending domain, whether to keep sending, rest it, or retire it. You compare each domain with its own history, not with the account average, and you do not retire a domain on noise.

## Step 1. Collect per-domain data

For each sending domain, per week (or per day at high volume):

- Emails sent (cold only, not warm-up), step 1 and follow-ups separately.
- Human replies and auto-replies (out-of-office, "received your message") as separate counts.
- Bounces by class, especially 5.7.x sender blocks.
- Seed test results with a neutral control, if any.
- Blocklist results for the domain and its IPs.
- Google Postmaster Tools data (spam rate, delivery errors) if the domain is verified there.
- Cumulative cold sends since the domain started sending.

Also note which campaigns and lists each domain sent to. A domain sending a weaker list is not burned.

## Step 2. Build the baseline

The baseline is the domain's own first 2 to 4 weeks of sending, or its best stable period. Compare only like with like: same campaign set, same step, similar lists.

## Step 3. Check the signals

| Signal | Burn sign |
|---|---|
| Step 1 human reply rate on an unchanged campaign set | down 50% or more vs the domain's own baseline, for 2+ weeks |
| Auto-reply rate | clear drop vs baseline. Auto-replies come from the recipient's server whatever the copy says, so they track delivery and react faster than human replies |
| Sender blocks (5.7.1, 5.7.x) | new or rising |
| Seed test with control | control lands in spam where it used to land in the inbox |
| Blocklists | new listing on Spamhaus DBL, SURBL or URIBL |
| Postmaster Tools | spam rate rising toward 0.3%, or delivery errors on reputation |

One signal alone is a flag. Two or more pointing the same way is a burn.

Rule out other causes first: a new list, a new copy, a change in follow-up share, a holiday week, or tiny samples. Do not judge a week with fewer than a few hundred sends from the domain; one or two replies up or down is noise at that size.

## Step 4. Track the service interval

Domains wear with cumulative cold volume. Log cumulative sends per domain and review each domain at a fixed interval (for example every ~1,500 cold sends) even when nothing looks wrong. Domains often decline gradually before anyone notices.

## Step 5. Decide

| Decision | When | What to do |
|---|---|---|
| Keep | no signal, or one weak signal with small samples | continue, re-check at the next interval |
| Rest | two signals, blocks without a listing, control still mostly inbox | stop cold sends for 1 to 3 weeks, keep warm-up running, restart with a fresh campaign and lower volume |
| Retire | control in spam on repeated tests, a major blocklist listing that returns after removal, or no recovery after rest | stop cold sending from the domain, keep it for replies until threads end, replace it |

## Output format

| Domain | Cum. sends | Baseline RR | Current RR | Auto-reply trend | Blocks | Seed (control) | Lists | Decision | Evidence |
|---|---|---|---|---|---|---|---|---|---|

Then: the domains to act on now, the replacement count needed to hold volume, and the next review date.

## Rules

- Compare a domain with its own baseline, never with the fleet average.
- Show sample sizes next to every rate.
- If evidence is weak, say "not enough data" and name the test that would settle it (usually a seed test with a control).

## Example

Input: outreach-example.com, cumulative 2,300 sends. Step 1 reply rate 1.2% in weeks 1 to 3, 0.5% in weeks 6 to 7 on the same campaigns. Auto-replies down by half. No blocklist listing. Control seed test: Google spam.

Output: Rest. Reply rate down 58% vs its own baseline on the same campaigns, auto-replies falling, control in spam on Google. Stop cold sends for 2 weeks, keep warm-up on, restart with a fresh campaign at half volume, run a control seed test before restarting.