A cold-email restart read at 48 hours: 41 sends to 31 agencies, 3 openers, all among the 10 contacted before the sequence fix, 0 of 21 contacted after it, 0 link clicks, 0 report loads, 0 replies, 0 bounces

Two populations, one campaign

The campaign is the one read at 24 hours in the linked post: one first email that names a client the agency lists on its own site, states one finding from a free AI-visibility check we ran on that client, links the report with a white-label invitation, and prices five full audits under the agency's name at $249; one follow-up declared for day 5. A configuration defect sent the 10 agencies contacted on 2026-09-02 their follow-up 9 minutes after the first email. The sequence was patched at 2026-09-03 10:57 UTC. Every lead contacted since has one email. So the campaign carries two populations and this read keeps them apart: P1, the 10 contacted before the fix (20 emails), and P2, the 21 contacted after it (21 emails).

The timing series

Sends by UTC day: 2026-09-02 20, 2026-09-03 20, 2026-09-04 1, under a limit of 20 a day across 2 inboxes. First emails by day: 2026-09-02 10, 2026-09-03 20, 2026-09-04 1. The 3 openers registered their latest open 0.2, 0.7, 9.8 hours after their first email. No P2 lead has registered an open; the oldest P2 leads are 24.3 h old and the median is 23 h, so on the P1 opener latencies (all under 10 hours) the P2 population has had the time in which P1's opens arrived.

What the split does and does not say

The two populations differ in more than the fix. P1 received two emails, the second a reply-threaded follow-up, and a second email doubles the chances of a tracking pixel firing. P1 went out on a Wednesday, P2 mostly on a Thursday afternoon. P1 is 10 leads and P2 is 21; at these sizes 3 of 10 against 0 of 21 is a difference worth reading tomorrow, not a conclusion today. What the read does settle: the fix held (every P2 lead has exactly one send, and the live delays are [5, 0] days), own bounce is 0% on 41 sends, and nobody has clicked the report link or loaded a report from any source in 47.9 hours.

Method

Four Instantly API v2 reads at 2026-09-04 13:52 UTC: the campaign object (sequence delays, daily limit, tracking flags), the campaign analytics row, every sent row with its step id and timestamp, and every lead row with its open, click and reply counts and the timestamp of its latest open. A lead's population is set by whether its first send predates the fix timestamp. Age is the read time minus the first send; opener latency is the latest open minus the first send (Instantly stores the latest open, not the first). Report loads are a HogQL count of PostHog $pageview events whose URL carries the campaign's UTM tag since the first send, with rows tagged utm_source=capture_probe (our own walks of the recipient URL shape, 2 rows on 2026-09-04) counted apart. Self-checks (11 of 11): sent rows equal the analytics count, lead rows equal the contacted and lead counts, the populations partition the leads, every P1 lead has two sends, every P2 lead has one send and no follow-up, no P2 follow-up sits inside the declared gap, the live delays carry the wait on step 1, openers agree between the analytics row and the lead rows, sends by day sum to the total, bounce is under the tripwire, and the artifact matches no address, domain or name pattern. Script: cold_restart_48h_cut.mjs; post text generated from the artifact by cold_restart_48h_post.mjs.

The dated reads, unchanged

1. 12026-09-05, the scored 48-hour give read, on P2: (A) at least 5 unique openers and 0 report loads means the report links, which sit behind the tracking redirect, are judged not to be opened from the email, and link tracking is switched off as a dated changed variable, read again 2026-09-09; (B) 0 openers on P2 points at inbox placement rather than the link, and per-lead opens, warmup scores and the tracking-domain reading are read before any field changes; (C) anything else changes nothing. Sending volume stays at 20 a day until that line prints.
2. 22026-09-08: proof of the fix by absence. No lead contacted from 2026-09-03 shows a follow-up inside 5 days of its first email. One such row pauses the campaign on sight.
3. 32026-09-09, one week: (A) at least one human reply, (B) zero replies with own bounce at or under 10% on at least 20 P2 sends, or (C) bounce over 10% or a spam complaint, which pauses the campaign. Scored on P1 and P2 separately.

Limits

• Opens are pixel-based. Clients that prefetch images register opens without a reader and clients that block images hide real ones; a two-email lead has two pixels. The opener counts are reported as Instantly prints them.
• Instantly stores a lead's latest open, not its first. The latencies above are upper bounds on time-to-first-open.
• 21 P2 leads at a median age of 23 h. The 24-hour and 48-hour marks on P2 are reached by 6 and 0 leads respectively; tomorrow's read is the one that scores.
• Report links pass through the tracking redirect. A click the redirect does not record and a load PostHog does not see are both invisible; the two counts agree at 0.
• Artifact: content/news/cold-restart-48h-cut-2026-09-04.json. Leads are numbered in first-send order; the file carries no address, name, domain, agency or client.